SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2001.02407 · ICLR · 2020

SPACE: Unsupervised Object-Oriented Scene Representation via Spatial Attention and Decomposition

Yi-Fu Wu, Skand Peri, Weihao Sun, Gautam Singh, Fei Deng, Jindong Jiang, Sungjin Ahn, Zhixuan Lin

arXiv · PDF · Open in the Atlas

Code that ran

36 of 51 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (25 of them constructed an object rather than computing a result; 3 honoured the contract we drafted, 0 violated it and 28 have no contract verdict; 5 failures of our instrument; 15 no recorded run), measured 2026-11-20; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2001.02407.json · badge: /api/ran/2001.02407.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
jinyangyuan/space community
Papers with Code archive
15 ran · 6 no recorded run
k4ntz/moc community
Papers with Code archive
10 ran · 6 no recorded run
zhixuan-lin/SPACE community
Papers with Code archive
10 ran · 3 no recorded run
sebamenabar/SPACE-Pytorch-Implementation community
Papers with Code archive
1 ran · 0 no recorded run
FunctionStatusWhere it lives
CompDecoder Ran
ran on a synthesized input · constructed an object; no result was computed
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("133f48b183df7d0b")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

CompDecoder Ran
ran on a synthesized input · constructed an object; no result was computed
zhixuan-lin/SPACE/src/model/space/space.py
code served (permissive licence) · get_code("f84d9362e8c0be0e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, CompDecoder: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from yacs.config import CfgNode

torch.manual_seed(0)

arch = CfgNode({'img_shape': (128, 128), 'G': 8, 'fg_sigma': 0.15, 'glimpse_size': 32, 'img_enc_dim_fg': 128, 'z_pres_dim': 1, 'z_depth_dim': 1, 'z_where_scale_dim': 2, 'z_where_shift_dim': 2, 'z_what_dim': 32, 'z_pres_start_step': 4000, 'z_pres_end_step': 10000, 'z_pres_start_value': 0.1, 'z_pres_end_value': 0.01, 'z_scale_mean_start_step': 10000, 'z_scale_mean_end_step': 20000, 'z_scale_mean_start_value': -1.0, 'z_scale_mean_end_value': -2.0, 'z_scale_std_value': 0.1, 'tau_start_step': 0, 'tau_end_step': 10000, 'tau_start_value': 2.5, 'tau_end_value': 2.5, 'boundary_loss': True, 'bl_off_step': 100000000, 'fix_alpha_steps': 0, 'fix_alpha_value': 0.1, 'K': 5, 'bg_sigma': 0.15, 'img_enc_dim_bg': 64, 'z_mask_dim': 32, 'z_comp_dim': 32, 'rnn_mask_hidden_dim': 64, 'rnn_mask_prior_hidden_dim': 64, 'predict_comp_hidden_dim': 64})

class SpatialBroadcast(nn.Module):
    """
    Broadcast a 1-D variable to 3-D, plus two coordinate dimensions
    """

    def __init__(self):
        nn.Module.__init__(self)

    def forward(self, x, width, height):
        """
        Broadcast a 1-D variable to 3-D, plus two coordinate dimensions
        
        :param x: (B, L)
        :param width: W
        :param height: H
        :return: (B, L + 2, W, H)
        """
        B, L = x.size()
        x = x[:, :, None, None]
        x = x.expand(B, L, width, height)
        xx = torch.linspace(-1, 1, width, device=x.device)
        yy = torch.linspace(-1, 1, height, device=x.device)
        yy, xx = torch.meshgrid((yy, xx))
        coords = torch.stack((xx, yy), dim=0)
        coords = coords[None].expand(B, 2, height, width)
        x = torch.cat((x, coords), dim=1)
        return x

class CompDecoder(nn.Module):
    """
    Decoder z_comp into component image
    """

    def __init__(self):
        nn.Module.__init__(self)
        self.spatial_broadcast = SpatialBroadcast()
        self.decoder = nn.Sequential(nn.Conv2d(arch.z_comp_dim + 2, 32, 3, 1), nn.BatchNorm2d(32), nn.ELU(), nn.Conv2d(32, 32, 3, 1), nn.BatchNorm2d(32), nn.ELU(), nn.Conv2d(32, 32, 3, 1), nn.BatchNorm2d(32), nn.ELU(), nn.Conv2d(32, 32, 3, 1), nn.BatchNorm2d(32), nn.ELU(), nn.Conv2d(32, 3, 1, 1))

    def forward(self, z_comp):
        """
        :param z_comp: (B, L)
        :return: component image (B, 3, H, W)
        """
        h, w = arch.img_shape
        z_comp = self.spatial_broadcast(z_comp, h + 8, w + 8)
        comp = self.decoder(z_comp)
        comp = torch.sigmoid(comp)
        return comp

ARGS = {}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, yacs — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type CompDecoder — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a CompDecoder output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
CompDecoderStrong Ran
ran on a synthesized input · constructed an object; no result was computed
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("1cbb4664acf8e289")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

CompDecoderStrong Ran
ran on a synthesized input · constructed an object; no result was computed
k4ntz/moc/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("0c8fd030940b81dd")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, k4ntz/moc, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

CompDecoderStrong Ran
ran on a synthesized input · constructed an object; no result was computed
zhixuan-lin/SPACE/src/model/space/space.py
code served (permissive licence) · get_code("453a7064dad87cfb")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, CompDecoderStrong: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from yacs.config import CfgNode

torch.manual_seed(0)

# Define the arch config as shown in the function signature
arch = CfgNode({'img_shape': (128, 128), 'G': 8, 'fg_sigma': 0.15, 'glimpse_size': 32, 'img_enc_dim_fg': 128, 'z_pres_dim': 1, 'z_depth_dim': 1, 'z_where_scale_dim': 2, 'z_where_shift_dim': 2, 'z_what_dim': 32, 'z_pres_start_step': 4000, 'z_pres_end_step': 10000, 'z_pres_start_value': 0.1, 'z_pres_end_value': 0.01, 'z_scale_mean_start_step': 10000, 'z_scale_mean_end_step': 20000, 'z_scale_mean_start_value': -1.0, 'z_scale_mean_end_value': -2.0, 'z_scale_std_value': 0.1, 'tau_start_step': 0, 'tau_end_step': 10000, 'tau_start_value': 2.5, 'tau_end_value': 2.5, 'boundary_loss': True, 'bl_off_step': 100000000, 'fix_alpha_steps': 0, 'fix_alpha_value': 0.1, 'K': 5, 'bg_sigma': 0.15, 'img_enc_dim_bg': 64, 'z_mask_dim': 32, 'z_comp_dim': 32, 'rnn_mask_hidden_dim': 64, 'rnn_mask_prior_hidden_dim': 64, 'predict_comp_hidden_dim': 64})

# CompDecoderStrong.__init__ takes no parameters beyond self
ARGS = {}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, yacs — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type CompDecoderStrong — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a CompDecoderStrong output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
CompEncoder Ran
ran on a synthesized input · constructed an object; no result was computed
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("a4ace48dedeef8fb")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

CompEncoder Ran
ran on a synthesized input · constructed an object; no result was computed
k4ntz/moc/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("60cecc1907723405")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, k4ntz/moc, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

CompEncoder Ran
ran on a synthesized input · constructed an object; no result was computed
zhixuan-lin/SPACE/src/model/space/space.py
code served (permissive licence) · get_code("bd2a5593f743092e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, CompEncoder: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import torch.nn.functional as F
from yacs.config import CfgNode

torch.manual_seed(0)

class Flatten(nn.Module):
    def forward(self, x):
        return x.view(x.size(0), -1)

arch = CfgNode({
    'img_shape': (128, 128),
    'G': 8,
    'fg_sigma': 0.15,
    'glimpse_size': 32,
    'img_enc_dim_fg': 128,
    'z_pres_dim': 1,
    'z_depth_dim': 1,
    'z_where_scale_dim': 2,
    'z_where_shift_dim': 2,
    'z_what_dim': 32,
    'z_pres_start_step': 4000,
    'z_pres_end_step': 10000,
    'z_pres_start_value': 0.1,
    'z_pres_end_value': 0.01,
    'z_scale_mean_start_step': 10000,
    'z_scale_mean_end_step': 20000,
    'z_scale_mean_start_value': -1.0,
    'z_scale_mean_end_value': -2.0,
    'z_scale_std_value': 0.1,
    'tau_start_step': 0,
    'tau_end_step': 10000,
    'tau_start_value': 2.5,
    'tau_end_value': 2.5,
    'boundary_loss': True,
    'bl_off_step': 100000000,
    'fix_alpha_steps': 0,
    'fix_alpha_value': 0.1,
    'K': 5,
    'bg_sigma': 0.15,
    'img_enc_dim_bg': 64,
    'z_mask_dim': 32,
    'z_comp_dim': 32,
    'rnn_mask_hidden_dim': 64,
    'rnn_mask_prior_hidden_dim': 64,
    'predict_comp_hidden_dim': 64
})

class CompEncoder(nn.Module):
    def __init__(self):
        nn.Module.__init__(self)
        embed_size = arch.img_shape[0] // 16
        self.enc = nn.Sequential(
            nn.Conv2d(4, 32, 3, 2, 1),
            nn.BatchNorm2d(32),
            nn.ELU(),
            nn.Conv2d(32, 32, 3, 2, 1),
            nn.BatchNorm2d(32),
            nn.ELU(),
            nn.Conv2d(32, 64, 3, 2, 1),
            nn.BatchNorm2d(64),
            nn.ELU(),
            nn.Conv2d(64, 64, 3, 2, 1),
            nn.BatchNorm2d(64),
            nn.ELU(),
            Flatten(),
            nn.Linear(64 * embed_size ** 2, arch.z_comp_dim * 2)
        )

    def forward(self, x):
        x = self.enc(x)
        z_comp_loc = x[:, :arch.z_comp_dim]
        z_comp_scale = F.softplus(x[:, arch.z_comp_dim:]) + 0.0001
        return (z_comp_loc, z_comp_scale)

ARGS = {}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, yacs — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type CompEncoder — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a CompEncoder output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
Flatten Ran
ran on a synthesized input
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("dca1bdd65a0d90e0")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

GlimpseDec Ran
ran on a synthesized input · constructed an object; no result was computed
zhixuan-lin/SPACE/src/model/space/space.py
code served (permissive licence) · get_code("351ffffa8d2c7190")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, GlimpseDec: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from yacs.config import CfgNode

torch.manual_seed(0)

arch = CfgNode({'img_shape': (128, 128), 'G': 8, 'fg_sigma': 0.15, 'glimpse_size': 32, 'img_enc_dim_fg': 128, 'z_pres_dim': 1, 'z_depth_dim': 1, 'z_where_scale_dim': 2, 'z_where_shift_dim': 2, 'z_what_dim': 32, 'z_pres_start_step': 4000, 'z_pres_end_step': 10000, 'z_pres_start_value': 0.1, 'z_pres_end_value': 0.01, 'z_scale_mean_start_step': 10000, 'z_scale_mean_end_step': 20000, 'z_scale_mean_start_value': -1.0, 'z_scale_mean_end_value': -2.0, 'z_scale_std_value': 0.1, 'tau_start_step': 0, 'tau_end_step': 10000, 'tau_start_value': 2.5, 'tau_end_value': 2.5, 'boundary_loss': True, 'bl_off_step': 100000000, 'fix_alpha_steps': 0, 'fix_alpha_value': 0.1, 'K': 5, 'bg_sigma': 0.15, 'img_enc_dim_bg': 64, 'z_mask_dim': 32, 'z_comp_dim': 32, 'rnn_mask_hidden_dim': 64, 'rnn_mask_prior_hidden_dim': 64, 'predict_comp_hidden_dim': 64})

class GlimpseDec(nn.Module):
    """Decoder z_what into reconstructed objects"""

    def __init__(self):
        super(GlimpseDec, self).__init__()
        self.dec = nn.Sequential(nn.Conv2d(arch.z_what_dim, 256, 1), nn.CELU(), nn.GroupNorm(16, 256), nn.Conv2d(256, 128 * 2 * 2, 1), nn.PixelShuffle(2), nn.CELU(), nn.GroupNorm(16, 128), nn.Conv2d(128, 128, 3, 1, 1), nn.CELU(), nn.GroupNorm(16, 128), nn.Conv2d(128, 128 * 2 * 2, 1), nn.PixelShuffle(2), nn.CELU(), nn.GroupNorm(16, 128), nn.Conv2d(128, 128, 3, 1, 1), nn.CELU(), nn.GroupNorm(16, 128), nn.Conv2d(128, 64 * 2 * 2, 1), nn.PixelShuffle(2), nn.CELU(), nn.GroupNorm(8, 64), nn.Conv2d(64, 64, 3, 1, 1), nn.CELU(), nn.GroupNorm(8, 64), nn.Conv2d(64, 32 * 2 * 2, 1), nn.PixelShuffle(2), nn.CELU(), nn.GroupNorm(8, 32), nn.Conv2d(32, 32, 3, 1, 1), nn.CELU(), nn.GroupNorm(8, 32), nn.Conv2d(32, 16 * 2 * 2, 1), nn.PixelShuffle(2), nn.CELU(), nn.GroupNorm(4, 16), nn.Conv2d(16, 16, 3, 1, 1), nn.CELU(), nn.GroupNorm(4, 16))
        self.dec_o = nn.Conv2d(16, 3, 3, 1, 1)
        self.dec_alpha = nn.Conv2d(16, 1, 3, 1, 1)

    def forward(self, x):
        """
        Decoder z_what into glimpse

        :param x: (B, D)
        :return:
            o_att: (B, 3, H, W)
            alpha_att: (B, 1, H, W)
        """
        x = self.dec(x.view(x.size(0), -1, 1, 1))
        o = torch.sigmoid(self.dec_o(x))
        alpha = torch.sigmoid(self.dec_alpha(x))
        return (o, alpha)

ARGS = {}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, yacs — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type GlimpseDec — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a GlimpseDec output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
GlimpseEncoder Ran
ran on a synthesized input
sebamenabar/SPACE-Pytorch-Implementation/utils/space_modules.py
pointer only (licence: NOASSERTION) · get_code("1f4684a24388ea46")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, sebamenabar/SPACE-Pytorch-Implementation (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

ImageEncoderBg Ran
ran on a synthesized input · constructed an object; no result was computed
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("730f01536a5df72a")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

ImageEncoderBg Ran
ran on a synthesized input · constructed an object; no result was computed
k4ntz/moc/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("50d23fc461c1ac05")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, k4ntz/moc, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

ImgEncoderFg Ran
ran on a synthesized input · constructed an object; no result was computed
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("8509f71e0541dfad")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

ImgEncoderFg Ran
ran on a synthesized input · constructed an object; no result was computed
k4ntz/moc/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("2b1ab6fda397d56c")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, k4ntz/moc, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

ImgEncoderFg Ran
ran on a synthesized input · constructed an object; no result was computed
zhixuan-lin/SPACE/src/model/space/space.py
code served (permissive licence) · get_code("2db58aa746031914")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, ImgEncoderFg: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from yacs.config import CfgNode

torch.manual_seed(0)

# Create the arch config as defined in the function
arch = CfgNode({
    'img_shape': (128, 128),
    'G': 8,
    'fg_sigma': 0.15,
    'glimpse_size': 32,
    'img_enc_dim_fg': 128,
    'z_pres_dim': 1,
    'z_depth_dim': 1,
    'z_where_scale_dim': 2,
    'z_where_shift_dim': 2,
    'z_what_dim': 32,
    'z_pres_start_step': 4000,
    'z_pres_end_step': 10000,
    'z_pres_start_value': 0.1,
    'z_pres_end_value': 0.01,
    'z_scale_mean_start_step': 10000,
    'z_scale_mean_end_step': 20000,
    'z_scale_mean_start_value': -1.0,
    'z_scale_mean_end_value': -2.0,
    'z_scale_std_value': 0.1,
    'tau_start_step': 0,
    'tau_end_step': 10000,
    'tau_start_value': 2.5,
    'tau_end_value': 2.5,
    'boundary_loss': True,
    'bl_off_step': 100000000,
    'fix_alpha_steps': 0,
    'fix_alpha_value': 0.1,
    'K': 5,
    'bg_sigma': 0.15,
    'img_enc_dim_bg': 64,
    'z_mask_dim': 32,
    'z_comp_dim': 32,
    'rnn_mask_hidden_dim': 64,
    'rnn_mask_prior_hidden_dim': 64,
    'predict_comp_hidden_dim': 64
})

# The ImgEncoderFg class requires no parameters in __init__
# It only uses the global arch variable and self initialization

ARGS = {}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, yacs — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ImgEncoderFg — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a ImgEncoderFg output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
MaskDecoder Ran
ran on a synthesized input · constructed an object; no result was computed
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("01c52fabe5c70752")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

MaskDecoder Ran
ran on a synthesized input
k4ntz/moc/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("a289c9c0362a7ac0")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, k4ntz/moc, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

MaskDecoder Ran
ran on a synthesized input · constructed an object; no result was computed
zhixuan-lin/SPACE/src/model/space/space.py
code served (permissive licence) · get_code("0784c8ca6cc7bd1a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, MaskDecoder: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from yacs.config import CfgNode

torch.manual_seed(0)

arch = CfgNode({'img_shape': (128, 128), 'G': 8, 'fg_sigma': 0.15, 'glimpse_size': 32, 'img_enc_dim_fg': 128, 'z_pres_dim': 1, 'z_depth_dim': 1, 'z_where_scale_dim': 2, 'z_where_shift_dim': 2, 'z_what_dim': 32, 'z_pres_start_step': 4000, 'z_pres_end_step': 10000, 'z_pres_start_value': 0.1, 'z_pres_end_value': 0.01, 'z_scale_mean_start_step': 10000, 'z_scale_mean_end_step': 20000, 'z_scale_mean_start_value': -1.0, 'z_scale_mean_end_value': -2.0, 'z_scale_std_value': 0.1, 'tau_start_step': 0, 'tau_end_step': 10000, 'tau_start_value': 2.5, 'tau_end_value': 2.5, 'boundary_loss': True, 'bl_off_step': 100000000, 'fix_alpha_steps': 0, 'fix_alpha_value': 0.1, 'K': 5, 'bg_sigma': 0.15, 'img_enc_dim_bg': 64, 'z_mask_dim': 32, 'z_comp_dim': 32, 'rnn_mask_hidden_dim': 64, 'rnn_mask_prior_hidden_dim': 64, 'predict_comp_hidden_dim': 64})

class MaskDecoder(nn.Module):
    """Decode z_mask into mask"""

    def __init__(self):
        super(MaskDecoder, self).__init__()
        self.dec = nn.Sequential(nn.Conv2d(arch.z_mask_dim, 256, 1), nn.CELU(), nn.GroupNorm(16, 256), nn.Conv2d(256, 256 * 4 * 4, 1), nn.PixelShuffle(4), nn.CELU(), nn.GroupNorm(16, 256), nn.Conv2d(256, 256, 3, 1, 1), nn.CELU(), nn.GroupNorm(16, 256), nn.Conv2d(256, 128 * 2 * 2, 1), nn.PixelShuffle(2), nn.CELU(), nn.GroupNorm(16, 128), nn.Conv2d(128, 128, 3, 1, 1), nn.CELU(), nn.GroupNorm(16, 128), nn.Conv2d(128, 64 * 4 * 4, 1), nn.PixelShuffle(4), nn.CELU(), nn.GroupNorm(8, 64), nn.Conv2d(64, 64, 3, 1, 1), nn.CELU(), nn.GroupNorm(8, 64), nn.Conv2d(64, 16 * 4 * 4, 1), nn.PixelShuffle(4), nn.CELU(), nn.GroupNorm(4, 16), nn.Conv2d(16, 16, 3, 1, 1), nn.CELU(), nn.GroupNorm(4, 16), nn.Conv2d(16, 16, 3, 1, 1), nn.CELU(), nn.GroupNorm(4, 16), nn.Conv2d(16, 1, 3, 1, 1))

    def forward(self, z_mask):
        """
        Decode z_mask into mask
        
        :param z_mask: (B, D)
        :return: mask: (B, 1, H, W)
        """
        B = z_mask.size(0)
        z_mask = z_mask.view(B, -1, 1, 1)
        mask = torch.sigmoid(self.dec(z_mask))
        return mask

ARGS = {}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, yacs — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type MaskDecoder — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a MaskDecoder output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
NumericalRelaxedBernoulli Ran
ran on a synthesized input · constructed an object; no result was computed
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("c0bdf5faa394c8c7")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

PredictComp Ran
ran on a synthesized input · constructed an object; no result was computed
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("29a57e7e4031100c")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

PredictComp Ran
ran on a synthesized input · constructed an object; no result was computed
k4ntz/moc/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("318a4a83d23ab99b")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, k4ntz/moc, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

PredictComp Ran
ran on a synthesized input · constructed an object; no result was computed
zhixuan-lin/SPACE/src/model/space/space.py
code served (permissive licence) · get_code("317c2513bc773e24")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, PredictComp: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import torch.nn.functional as F
from yacs.config import CfgNode

torch.manual_seed(0)

arch = CfgNode({
    'img_shape': (128, 128),
    'G': 8,
    'fg_sigma': 0.15,
    'glimpse_size': 32,
    'img_enc_dim_fg': 128,
    'z_pres_dim': 1,
    'z_depth_dim': 1,
    'z_where_scale_dim': 2,
    'z_where_shift_dim': 2,
    'z_what_dim': 32,
    'z_pres_start_step': 4000,
    'z_pres_end_step': 10000,
    'z_pres_start_value': 0.1,
    'z_pres_end_value': 0.01,
    'z_scale_mean_start_step': 10000,
    'z_scale_mean_end_step': 20000,
    'z_scale_mean_start_value': -1.0,
    'z_scale_mean_end_value': -2.0,
    'z_scale_std_value': 0.1,
    'tau_start_step': 0,
    'tau_end_step': 10000,
    'tau_start_value': 2.5,
    'tau_end_value': 2.5,
    'boundary_loss': True,
    'bl_off_step': 100000000,
    'fix_alpha_steps': 0,
    'fix_alpha_value': 0.1,
    'K': 5,
    'bg_sigma': 0.15,
    'img_enc_dim_bg': 64,
    'z_mask_dim': 32,
    'z_comp_dim': 32,
    'rnn_mask_hidden_dim': 64,
    'rnn_mask_prior_hidden_dim': 64,
    'predict_comp_hidden_dim': 64
})

class PredictComp(nn.Module):
    """
    Predict component latents given mask latent
    """

    def __init__(self):
        nn.Module.__init__(self)
        self.mlp = nn.Sequential(
            nn.Linear(arch.z_mask_dim, arch.predict_comp_hidden_dim),
            nn.ELU(),
            nn.Linear(arch.predict_comp_hidden_dim, arch.predict_comp_hidden_dim),
            nn.ELU(),
            nn.Linear(arch.predict_comp_hidden_dim, arch.z_comp_dim * 2)
        )

    def forward(self, h):
        """
        :param h: (B, D) hidden state from rnn_mask
        :return:
            z_comp_loc: (B, D)
            z_comp_scale: (B, D)
        """
        x = self.mlp(h)
        z_comp_loc = x[:, :arch.z_comp_dim]
        z_comp_scale = F.softplus(x[:, arch.z_comp_dim:]) + 0.0001
        return (z_comp_loc, z_comp_scale)


# Create input tensor: (batch_size=2, z_mask_dim=32)
h = torch.randn(2, arch.z_mask_dim)

ARGS = {}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, yacs — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PredictComp — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PredictComp output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
PredictMask Ran
ran on a synthesized input · constructed an object; no result was computed
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("d7466602b09144bd")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

PredictMask Ran
ran on a synthesized input · constructed an object; no result was computed
zhixuan-lin/SPACE/src/model/space/space.py
code served (permissive licence) · get_code("9b80c13dd98959d4")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, PredictMask: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import torch.nn.functional as F
from yacs.config import CfgNode

torch.manual_seed(0)

arch = CfgNode({'img_shape': (128, 128), 'G': 8, 'fg_sigma': 0.15, 'glimpse_size': 32, 'img_enc_dim_fg': 128, 'z_pres_dim': 1, 'z_depth_dim': 1, 'z_where_scale_dim': 2, 'z_where_shift_dim': 2, 'z_what_dim': 32, 'z_pres_start_step': 4000, 'z_pres_end_step': 10000, 'z_pres_start_value': 0.1, 'z_pres_end_value': 0.01, 'z_scale_mean_start_step': 10000, 'z_scale_mean_end_step': 20000, 'z_scale_mean_start_value': -1.0, 'z_scale_mean_end_value': -2.0, 'z_scale_std_value': 0.1, 'tau_start_step': 0, 'tau_end_step': 10000, 'tau_start_value': 2.5, 'tau_end_value': 2.5, 'boundary_loss': True, 'bl_off_step': 100000000, 'fix_alpha_steps': 0, 'fix_alpha_value': 0.1, 'K': 5, 'bg_sigma': 0.15, 'img_enc_dim_bg': 64, 'z_mask_dim': 32, 'z_comp_dim': 32, 'rnn_mask_hidden_dim': 64, 'rnn_mask_prior_hidden_dim': 64, 'predict_comp_hidden_dim': 64})

class PredictMask(nn.Module):
    """
    Predict z_mask given states from rnn. Used in inference
    """

    def __init__(self):
        nn.Module.__init__(self)
        self.fc = nn.Linear(arch.rnn_mask_hidden_dim, arch.z_mask_dim * 2)

    def forward(self, h):
        """
        Predict z_mask given states from rnn. Used in inference
    
        :param h: hidden state from rnn_mask
        :return:
            z_mask_loc: (B, D)
            z_mask_scale: (B, D)
        
        """
        x = self.fc(h)
        z_mask_loc = x[:, :arch.z_mask_dim]
        z_mask_scale = F.softplus(x[:, arch.z_mask_dim:]) + 0.0001
        return (z_mask_loc, z_mask_scale)

# Create an instance of PredictMask
predict_mask = PredictMask()

# Create a sample hidden state tensor with batch size 2 and hidden dim 64
h = torch.randn(2, arch.rnn_mask_hidden_dim)

ARGS = {}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, yacs — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PredictMask — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PredictMask output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
ZWhatEnc Ran
ran on a synthesized input · constructed an object; no result was computed
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("a5f8ccf2364b391c")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

ZWhatEnc Ran
ran on a synthesized input · constructed an object; no result was computed
k4ntz/moc/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("a0176b778c024a84")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, k4ntz/moc, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

ZWhatEnc Ran
ran on a synthesized input · constructed an object; no result was computed
zhixuan-lin/SPACE/src/model/space/space.py
code served (permissive licence) · get_code("22f02c42a137247f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, ZWhatEnc: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from yacs.config import CfgNode

torch.manual_seed(0)

arch = CfgNode({'img_shape': (128, 128), 'G': 8, 'fg_sigma': 0.15, 'glimpse_size': 32, 'img_enc_dim_fg': 128, 'z_pres_dim': 1, 'z_depth_dim': 1, 'z_where_scale_dim': 2, 'z_where_shift_dim': 2, 'z_what_dim': 32, 'z_pres_start_step': 4000, 'z_pres_end_step': 10000, 'z_pres_start_value': 0.1, 'z_pres_end_value': 0.01, 'z_scale_mean_start_step': 10000, 'z_scale_mean_end_step': 20000, 'z_scale_mean_start_value': -1.0, 'z_scale_mean_end_value': -2.0, 'z_scale_std_value': 0.1, 'tau_start_step': 0, 'tau_end_step': 10000, 'tau_start_value': 2.5, 'tau_end_value': 2.5, 'boundary_loss': True, 'bl_off_step': 100000000, 'fix_alpha_steps': 0, 'fix_alpha_value': 0.1, 'K': 5, 'bg_sigma': 0.15, 'img_enc_dim_bg': 64, 'z_mask_dim': 32, 'z_comp_dim': 32, 'rnn_mask_hidden_dim': 64, 'rnn_mask_prior_hidden_dim': 64, 'predict_comp_hidden_dim': 64})

ARGS = {}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, yacs — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ZWhatEnc — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a ZWhatEnc output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_boundary_kernel Ran
ran; our contract draft was wrong, not the code
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("cec75c84bc88045c")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_boundary_kernel_new Ran
ran, honoured the contract we drafted
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("5cefba6de921f217")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

inverse_spatial_transform Ran
ran; our contract draft was wrong, not the code
k4ntz/moc/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("0a1e474f8c6f9235")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, k4ntz/moc, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

linear_annealing Ran
ran, honoured the contract we drafted
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("c2ecfa7b95f02508")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

linear_annealing Ran
ran, honoured the contract we drafted
k4ntz/moc/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("bfca6ead307936c8")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, k4ntz/moc, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

spatial_transform Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("8582eaff32ef7fff")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

spatial_transform Ran
ran; our contract draft was wrong, not the code
k4ntz/moc/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("f2b3c5bce3661feb")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, k4ntz/moc, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

spatial_transform Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zhixuan-lin/SPACE/src/model/space/space.py
code served (permissive licence) · get_code("5da93a38a1779bcb")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 4, dtype family float, shape (2, 1, 5, 5) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn.functional as F

torch.manual_seed(0)

# Construct arguments for spatial_transform function
batch_size = 2
image_height = 4
image_width = 3
attn_window = 5
num_channels = 1

# Create a small batch of images
image = torch.randn(batch_size, num_channels, image_height, image_width)

# Create z_where parameters: [scale_x, scale_y, tx, ty]
z_where = torch.randn(batch_size, 4)

# Output dimensions for the spatial transformation
out_dims = torch.Size([batch_size, num_channels, attn_window, attn_window])

# Boolean flag for inverse transformation
inverse = False

ARGS = {
    'image': image,
    'z_where': z_where,
    'out_dims': out_dims,
    'inverse': inverse
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 1, 5, 5) — shape (2, 1, 5, 5) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
CompDecoder Not yet run
unverified (harvested, no recorded run)
k4ntz/moc/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("dc8e1056e4079750")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, k4ntz/moc, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

GlimpseDec Not yet run
unverified (harvested, no recorded run)
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("86d183f89f3bbd32")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

GlimpseDec Not yet run
unverified (harvested, no recorded run)
k4ntz/moc/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("af315f04ad9a060f")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, k4ntz/moc, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

PredictMask Not yet run
unverified (harvested, no recorded run)
k4ntz/moc/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("61c3acd8b00e59e6")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, k4ntz/moc, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Space Not yet run
unverified (harvested, no recorded run)
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("394e067bedabb297")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Space Not yet run
unverified (harvested, no recorded run)
k4ntz/moc/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("6abb28fbe1178f97")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, k4ntz/moc, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Space Not yet run
unverified (harvested, no recorded run)
zhixuan-lin/SPACE/src/model/space/space.py
code served (permissive licence) · get_code("12db70167d3291df")
How it ran

no invocation recorded.

SpaceBg Not yet run
unverified (harvested, no recorded run)
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("a1fdc89a469eb273")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

SpaceBg Not yet run
unverified (harvested, no recorded run)
k4ntz/moc/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("3c0f8ef59ad36050")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, k4ntz/moc, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

SpaceBg Not yet run
unverified (harvested, no recorded run)
zhixuan-lin/SPACE/src/model/space/space.py
code served (permissive licence) · get_code("bcfb9765a024bd2d")
How it ran

no invocation recorded.

SpaceFg Not yet run
unverified (harvested, no recorded run)
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("c3414acc0ae26193")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

SpaceFg Not yet run
unverified (harvested, no recorded run)
k4ntz/moc/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("e1a39d7201652ecc")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, k4ntz/moc, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

SpaceFg Not yet run
unverified (harvested, no recorded run)
zhixuan-lin/SPACE/src/model/space/space.py
code served (permissive licence) · get_code("4c193d00ee677ab4")
How it ran

no invocation recorded.

SpatialBroadcast Not yet run
unverified (harvested, no recorded run)
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("fa03a464466fad92")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

kl_divergence_bern_bern Not yet run
unverified (harvested, no recorded run)
jinyangyuan/space/src/model/space/space.py
pointer only (licence: GPL-3.0) · get_code("3fc240625d53914a")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, jinyangyuan/space, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Also present in other papers' code

Of 51 samples lifted from this paper's linked repositories, 51 carry a function identity and 0 do not. Of the 40 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 1 later) and 39 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
Flatten
brick of this paper's assembly: helper
15 other papers
7 dated earlier, 8 later · 14 only as an assembly brick
1411.7766 (2014-11)
Deep Learning Face Attributes in the Wild *
1503.03832 (2015-03)
FaceNet: A Unified Embedding for Face Recognition and Clustering
1505.04597 (2015-05)
U-Net: Convolutional Networks for Biomedical Image Segmentation
and 12 more not named here

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-20.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

The ability to decompose complex multi-object scenes into meaningful abstractions like objects is fundamental to achieve higher-level cognition. Previous approaches for unsupervised object-oriented scene representation learning are either based on spatial-attention or scene-mixture approaches and limited in scalability which is a main obstacle towards modeling real-world scenes. In this paper, we propose a generative latent variable model, called SPACE, that provides a unified probabilistic modeling framework that combines the best of spatial-attention and scene-mixture approaches. SPACE can explicitly provide factorized object representations for foreground objects while also decomposing background segments of complex morphology. Previous models are good at either of these, but not both. SPACE also resolves the scalability problems of previous methods by incorporating parallel spatial-attention and thus is applicable to scenes with a large number of objects without performance degradations. We show through experiments on Atari and 3D-Rooms that SPACE achieves the above properties consistently in comparison to SPAIR, IODINE, and GENESIS. Results of our experiments can be found on our project website: https://sites.google.com/view/space-project-page

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2001.02407")
get_code_for_paper("2001.02407")
have("2001.02407")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2001.02407.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2001.02407.svg)](https://syntology.ai/paper/2001.02407)

Connect an agent — have() is free.