SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2109.03891 · 2021

SORNet: Spatial Object-Centric Representations for Sequential Manipulation

arXiv · PDF · Open in the Atlas

Code that ran

5 of 7 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 5 have no contract verdict; 0 failures of our instrument; 2 no recorded run), measured 2026-11-22; 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/2109.03891.json · badge: /api/ran/2109.03891.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
wentaoyuan/sornet canonical (classifier label)
classifier basis: census
5 ran · 2 no recorded run
FunctionStatusWhere it lives
build_predicates Ran
ran on a synthesized input
wentaoyuan/sornet/datasets.py
code served (permissive licence) · get_code("0c414c16502ebb56")
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 list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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 numpy as np

# Analyze the function:
# - objects: used in list comprehensions with `for obj in objects` and `np.roll(objects, -i)`
#   This suggests it should be an iterable (list, array, etc.) of strings or identifiers
# - unary: used as `for pred in unary`, where each pred is formatted with `pred % obj`
#   This suggests a list of format strings with one %s placeholder
# - binary: used as `for pred in binary`, where each pred is formatted with `pred % (o1, o2)`
#   This suggests a list of format strings with two %s placeholders

# Build minimal valid arguments:
objects = ["obj_a", "obj_b", "obj_c"]
unary = ["pred_unary_%s", "color_%s"]
binary = ["pred_rel_%s_%s", "near_%s_%s"]

ARGS = {
    "objects": objects,
    "unary": unary,
    "binary": binary
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
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
calc_acc Ran
ran on a synthesized input
wentaoyuan/sornet/train_leonardo.py
code served (permissive licence) · get_code("340a4f2d4088b73e")
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 1, dtype family float, shape (4,) 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 as nn

torch.manual_seed(0)

# Construct arguments for calc_acc function
# The function expects:
# - data: tuple of (img, obj_patches, gripper, target)
# - logits: tensor that will be compared with target
# - pred_types: appears unused in function body

# Create small tensors with distinct sizes
batch_size = 2
num_classes = 4

# img: small image tensor
img = torch.randn(batch_size, 3, 8, 8)

# obj_patches: object patches tensor
obj_patches = torch.randn(batch_size, 5, 16)

# gripper: gripper state tensor
gripper = torch.randn(batch_size, 2)

# target: binary target tensor (will be compared with (logits > 0))
target = torch.randint(0, 2, (batch_size, num_classes)).float()

# Combine into data tuple
data = (img, obj_patches, gripper, target)

# logits: tensor with shape (batch_size, num_classes) to compare with target
logits = torch.randn(batch_size, num_classes)

# pred_types: unused in function body, can be any value
pred_types = None

ARGS = {
    "data": data,
    "logits": logits,
    "pred_types": pred_types
}
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 1, dtype family float, shape (4,) — shape (4,) 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
calc_dist Ran
ran on a synthesized input
wentaoyuan/sornet/train_regression.py
code served (permissive licence) · get_code("e42cc40dfb34d87e")
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 0, dtype family float, shape () 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 as nn

torch.manual_seed(0)

# Build the data tuple (img, obj_patches, target)
# target shape needs to work with logits after reshape
# logits will be reshaped to (batch_size, target.shape[2], -1) then transposed to (batch_size, -1, target.shape[2])
# So logits has shape (batch_size, height, width, channels) initially
# After reshape: (batch_size, target.shape[2], height*width)
# After transpose: (batch_size, height*width, target.shape[2])
# target should match this shape after transpose

batch_size = 2
height = 3
width = 4
target_dim = 5

# Create target with shape (batch_size, height*width, target_dim)
target = torch.randn(batch_size, height * width, target_dim)

# Create img and obj_patches as dummy tensors
img = torch.randn(batch_size, 3, 32, 32)
obj_patches = torch.randn(batch_size, 8, 16, 16)

data = (img, obj_patches, target)

# Create logits with shape (batch_size, height, width, target_dim)
# This will reshape to (batch_size, target_dim, height*width)
# Then transpose to (batch_size, height*width, target_dim)
logits = torch.randn(batch_size, height, width, target_dim)

# pred_types is not used in the function body, so it can be any value
pred_types = None

ARGS = {
    "data": data,
    "logits": logits,
    "pred_types": pred_types
}
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 0, dtype family float, shape () — shape () 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
step Ran
ran on a synthesized input
wentaoyuan/sornet/train_clevr.py
code served (permissive licence) · get_code("70cbeab60da02673")
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 tuple. 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

torch.manual_seed(0)

# Define a simple model that takes img and obj_patches and returns emb and attn
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 5)
    
    def forward(self, img, obj_patches):
        # img shape: (batch, 4)
        # obj_patches shape: (batch, 3, 4)
        # Combine and process
        combined = img.unsqueeze(1) + obj_patches  # (batch, 3, 4)
        emb = self.linear(combined.mean(dim=1))  # (batch, 5)
        attn = torch.ones(2, 3)  # (batch, num_patches)
        return emb, attn

# Define a simple head that takes emb and returns logits
class SimpleHead(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(5, 2)
    
    def forward(self, emb):
        # emb shape: (batch, 5)
        logits = self.linear(emb)  # (batch, 2)
        return logits

# Create model and head instances
model = SimpleModel()
head = SimpleHead()

# Create data: (img, obj_patches, target, mask)
batch_size = 2
img = torch.randn(batch_size, 4)  # (batch, 4)
obj_patches = torch.randn(batch_size, 3, 4)  # (batch, num_patches, 4)
target = torch.randint(0, 2, (batch_size, 2)).float()  # (batch, 2)
mask = torch.ones(batch_size, 2)  # (batch, 2)

data = (img, obj_patches, target, mask)

ARGS = {
    'data': data,
    'model': model,
    'head': head
}
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 tuple — a tuple output has no array shape of its own; none is 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
step Ran
ran on a synthesized input
wentaoyuan/sornet/train_regression.py
code served (permissive licence) · get_code("565a7402728223c4")
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 tuple. 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

torch.manual_seed(0)

# Define a minimal model that matches how it's called in step()
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear_emb = nn.Linear(12, 8)  # processes concatenated img + obj_patches
        self.linear_attn = nn.Linear(12, 4)
    
    def forward(self, img, obj_patches):
        # img: (batch, 4), obj_patches: (batch, 8)
        combined = torch.cat([img, obj_patches], dim=1)  # (batch, 12)
        emb = self.linear_emb(combined)  # (batch, 8)
        attn = self.linear_attn(combined)  # (batch, 4)
        return emb, attn

# Define a minimal head that matches how it's called in step()
class SimpleHead(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(8, 6)  # emb is (batch, 8), output logits (batch, 6)
    
    def forward(self, emb):
        return self.linear(emb)

# Create model and head instances
model = SimpleModel()
head = SimpleHead()

# Create data tuple
# img: (batch=2, 4)
img = torch.randn(2, 4)
# obj_patches: (batch=2, 8)
obj_patches = torch.randn(2, 8)
# target: must reshape to (batch, 6) after transpose(1,2)
# target.transpose(1, 2).reshape(logits.shape[0], -1) where logits is (2, 6)
# So target needs to be (batch, ?, ?) such that after transpose and reshape it becomes (2, 6)
# If target is (2, 2, 3), transpose(1,2) gives (2, 3, 2), reshape(2, -1) gives (2, 6) ✓
target = torch.randn(2, 2, 3)

data = (img, obj_patches, target)

ARGS = {
    "data": data,
    "model": model,
    "head": head
}
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 tuple — a tuple output has no array shape of its own; none is 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
calc_acc Not yet run
unverified (harvested, no recorded run)
wentaoyuan/sornet/train_clevr.py
code served (permissive licence) · get_code("baad60b39603aa26")
How it ran

no invocation recorded.

step Not yet run
unverified (harvested, no recorded run)
wentaoyuan/sornet/train_leonardo.py
code served (permissive licence) · get_code("263ba8b7b267c3e4")
How it ran

no invocation recorded.

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 7 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.

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 2026-11-22.

Repositories linked to this paper

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

Abstract

Sequential manipulation tasks require a robot to perceive the state of an environment and plan a sequence of actions leading to a desired goal state. In such tasks, the ability to reason about spatial relations among object entities from raw sensor inputs is crucial in order to determine when a task has been completed and which actions can be executed. In this work, we propose SORNet (Spatial Object-Centric Representation Network), a framework for learning object-centric representations from RGB images conditioned on a set of object queries, represented as image patches called canonical object views. With only a single canonical view per object and no annotation, SORNet generalizes zero-shot to object entities whose shape and texture are both unseen during training. We evaluate SORNet on various spatial reasoning tasks such as spatial relation classification and relative direction regression in complex tabletop manipulation scenarios and show that SORNet significantly outperforms baselines including state-of-the-art representation learning techniques. We also demonstrate the application of the representation learned by SORNet on visual-servoing and task planning for sequential manipulation on a real robot.

For agents

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

get_harvested_code_for_paper("2109.03891")
get_code_for_paper("2109.03891")
have("2109.03891")

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

curl https://syntology.ai/api/ran/2109.03891.json

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

[![Syntology run record](https://syntology.ai/api/ran/2109.03891.svg)](https://syntology.ai/paper/2109.03891)

Connect an agent — have() is free.