SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2502.18364 · CVPR · 2025

ART: Anonymous Region Transformer for Variable Multi-Layer Transparent Image Generation

Gao Huang, Lijuan Wang, Xiu Li, Zhouhui Lian, Dong Chen, Yuhui Yuan, Yifan Pu, Baining Guo, Lin Liang, Ji Li, Yiming Zhao, Ruihong Yin, and 5 more

arXiv · PDF · Open in the Atlas

Code that ran

13 of 14 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 11 have no contract verdict; 2 failures of our instrument; 1 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/2502.18364.json · badge: /api/ran/2502.18364.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
microsoft/art-msra community (classifier label)
classifier basis not recorded on the link
13 ran · 1 no recorded run
FunctionStatusWhere it lives
FormulateInput Ran
ran on a synthesized input
microsoft/art-msra/layout_planner/inference_layout.py
code served (permissive licence) · get_code("07f3e48fb520f32b")
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 dict. 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)

# The function takes a single parameter 'intention' which is a string
# It creates a dict with 'wholecaption' set to the intention string
# and 'layout' set to an empty list
# The function doesn't actually use the return value in any complex way,
# so we just need to provide a valid string input

ARGS = {
    "intention": "A beautiful landscape with mountains and a sunset"
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
accuracy Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
microsoft/art-msra/layout_planner/training/utils.py
code served (permissive licence) · get_code("ed5ed903e9e24d08")
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 2, 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 torch

torch.manual_seed(0)

# Construct valid arguments for the accuracy function
# output: tensor of shape (batch_size, num_classes) with prediction logits/scores
# target: tensor of shape (batch_size,) with class indices
# padding: integer value used to mask certain positions
# topk: tuple of integers specifying which top-k accuracies to compute

batch_size = 2
num_classes = 4
num_elements = 3

# output: prediction scores/logits, shape (batch_size, num_classes)
output = torch.randn(batch_size, num_classes)

# target: ground truth class indices, shape (batch_size,)
# Use values in range [0, num_classes) with some padding values
target = torch.tensor([0, 2], dtype=torch.long)

# padding: special value indicating positions to ignore
padding = -1

# topk: tuple of k values to compute accuracy for
topk = (1, 2)

ARGS = {
    "output": output,
    "target": target,
    "padding": padding,
    "topk": topk
}
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 list, ndim 2, 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
batch_purity Ran
ran on a synthesized input
microsoft/art-msra/layout_planner/training/trainer_layout.py
code served (permissive licence) · get_code("b6ae4c2e9a4721f3")
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 torch

torch.manual_seed(0)

class MockTokenizer:
    def batch_decode(self, input_ids):
        # Return a list of strings matching batch size
        batch_size = input_ids.shape[0] if hasattr(input_ids, 'shape') else len(input_ids)
        return [f'<s> hello> "world</s>' for _ in range(batch_size)]

input_ids = torch.tensor([[1, 2, 3, 4], [5, 6, 7, 8]])
tokenizer = MockTokenizer()

ARGS = {
    'input_ids': input_ids,
    'tokenizer': tokenizer,
}
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 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
crop_each_layer Ran
ran on a synthesized input
microsoft/art-msra/multi_layer_gen/custom_model_transp_vae.py
code served (permissive licence) · get_code("28e0f66d8fa6cea8")
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 pos_embedding model that returns the expected structure
class PosEmbedding(nn.Module):
    def forward(self, ids):
        # ids shape: (seq_len, 3)
        # Should return tuple of (cos_embed, sin_embed)
        # Each should be reshapeable to (H, W, 64)
        seq_len = ids.shape[0]
        cos_embed = torch.randn(seq_len, 64)
        sin_embed = torch.randn(seq_len, 64)
        return (cos_embed, sin_embed)

# Create small test tensors
batch_size = 2
num_layers = 3
height = 16  # H dimension
width = 12   # W dimension
channels = 4

# hidden_states: (batch, num_layers, height, width)
hidden_states = torch.randn(batch_size, num_layers, height, width)

# use_layers: list indicating which layer each element uses
use_layers = [0, 1, 2]

# list_layer_box: list of bounding boxes (x1, y1, x2, y2) or None
# Each box should be divisible by 8 for the // 8 operation
list_layer_box = [
    (0, 0, 16, 16),      # Layer 0: box that covers some area
    (8, 8, 24, 24),      # Layer 1: another box
    None,                # Layer 2: no box (will be skipped)
]

H = 8   # Height for position embedding (different from hidden_states height)
W = 6   # Width for position embedding (different from hidden_states width)

pos_embedding = PosEmbedding()

ARGS = {
    "hidden_states": hidden_states,
    "use_layers": use_layers,
    "list_layer_box": list_layer_box,
    "H": H,
    "W": W,
    "pos_embedding": pos_embedding,
}
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
encode_prompt Ran
ran on a synthesized input
microsoft/art-msra/multi_layer_gen/custom_pipeline.py
code served (permissive licence) · get_code("06d07c5e65ea4167")
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)

# Stand-in config class for CLIP text encoder
class CLIPConfig:
    def __init__(self):
        self.max_position_embeddings = 77

# Stand-in CLIP tokenizer
class CLIPTokenizer:
    def __call__(self, prompt, padding=None, max_length=None, truncation=None, 
                 return_overflowing_tokens=None, return_length=None, return_tensors=None):
        if isinstance(prompt, str):
            prompt = [prompt]
        batch_size = len(prompt)
        # Return a dict-like object with input_ids
        class TokenizerOutput:
            def __init__(self, ids):
                self.input_ids = ids
        return TokenizerOutput(torch.zeros((batch_size, max_length), dtype=torch.long))

# Stand-in T5 tokenizer
class T5Tokenizer:
    def __call__(self, prompt, padding=None, max_length=None, truncation=None,
                 return_length=None, return_overflowing_tokens=None, return_tensors=None):
        if isinstance(prompt, str):
            prompt = [prompt]
        batch_size = len(prompt)
        class TokenizerOutput:
            def __init__(self, ids):
                self.input_ids = ids
        return TokenizerOutput(torch.zeros((batch_size, max_length), dtype=torch.long))

# Stand-in CLIP text encoder
class CLIPTextEncoder(nn.Module):
    def __init__(self):
        super().__init__()
        self.config = CLIPConfig()
        self.embed_dim = 4  # Small embedding dimension
        self.pooler = nn.Linear(self.embed_dim, self.embed_dim)
        
    def forward(self, input_ids, output_hidden_states=False):
        batch_size, seq_len = input_ids.shape
        # Return embeddings with pooler_output
        hidden_states = torch.randn(batch_size, seq_len, self.embed_dim)
        pooled = self.pooler(hidden_states[:, 0, :])  # Take first token for pooling
        
        class EncoderOutput:
            def __init__(self, pooler_output):
                self.pooler_output = pooler_output
        
        return EncoderOutput(pooled)

# Stand-in T5 text encoder
class T5TextEncoder(nn.Module):
    def __init__(self):
        super().__init__()
        self.embed_dim = 5  # Different dim than CLIP for distinctness
        self.seq_len = 10
        
    def forward(self, input_ids, output_hidden_states=False):
        batch_size, seq_len = input_ids.shape
        # T5 returns tuple where first element is hidden states
        hidden_states = torch.randn(batch_size, seq_len, self.embed_dim)
        return (hidden_states,)

# Create instances
tokenizer_1 = CLIPTokenizer()
tokenizer_2 = T5Tokenizer()

text_encoder_1 = CLIPTextEncoder()
text_encoder_1.device = torch.device('cpu')
text_encoder_1.dtype = torch.float32

text_encoder_2 = T5TextEncoder()
text_encoder_2.device = torch.device('cpu')
text_encoder_2.dtype = torch.float32

# Build arguments for encode_prompt
ARGS = {
    'tokenizers': (tokenizer_1, tokenizer_2),
    'text_encoders': (text_encoder_1, text_encoder_2),
    'prompt': ['a dog', 'a cat'],
    'prompt_2': ['a happy dog', 'a sleepy cat'],
    'num_images_per_prompt': 2,
    'max_sequence_length': 512,
}
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
get_quantizer Ran
ran on a synthesized input
microsoft/art-msra/layout_planner/training/datasets/quantizer.py
code served (permissive licence) · get_code("c345a9eaf4ac73c4")
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 QuantizerV4. 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 numpy as np

ARGS = {
    'version': 'v4',
    'quant': True,
    'simplify_json': False,
    'width': 1024,
    'height': 1024,
}
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 QuantizerV4 — a QuantizerV4 output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a QuantizerV4 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
kmp_preprocess Ran
ran on a synthesized input
microsoft/art-msra/layout_planner/models/modeling_layout.py
code served (permissive licence) · get_code("00462e6910b34cb9")
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 int; 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 torch

torch.manual_seed(0)

# kmp_preprocess takes a pattern parameter
# It treats pattern as a sequence that supports:
# - len(pattern)
# - indexing with pattern[i]
# - iteration

# A simple list works perfectly for this function
pattern = [1, 2, 3, 2, 3, 1, 2, 3]

ARGS = {
    "pattern": pattern
}
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 list, ndim 1, dtype family int — 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
kmp_search Ran
ran on a synthesized input
microsoft/art-msra/layout_planner/models/modeling_layout.py
code served (permissive licence) · get_code("f150a306a653ecb2")
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 int; 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.

# KMP search function setup
# The function searches for a pattern in text using the Knuth-Morris-Pratt algorithm
# Both 'text' and 'pattern' are sequences that support indexing and len()

# Simple list-based sequences work fine
text = ['a', 'b', 'c', 'a', 'b', 'a', 'b', 'c']
pattern = ['a', 'b', 'c']

ARGS = {
    'text': text,
    'pattern': pattern
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — 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 int — 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
l2_loss Ran
ran; our contract draft was wrong, not the code
microsoft/art-msra/layout_planner/training/trainer_layout.py
code served (permissive licence) · get_code("61490145cb48f028")
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 2, dtype family float, shape (2, 3) 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

torch.manual_seed(0)

# Create two tensors with shape (N, T, D)
# N=2 (batch size), T=3 (time steps), D=4 (feature dimension)
u = torch.randn(2, 3, 4)
v = torch.randn(2, 3, 4)

ARGS = {
    "u": u,
    "v": v,
}
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 2, dtype family float, shape (2, 3) — shape (2, 3) 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
layout_collate_fn Ran
ran on a synthesized input
microsoft/art-msra/layout_planner/training/datasets/layoutnew_dataset.py
code served (permissive licence) · get_code("f4d9f77d7e39090d")
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 dict. 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

torch.manual_seed(0)

# The function expects batch to be a list of dicts
# Each dict should have 'labels' key with a tensor value
# Optionally, each dict can have an 'index' key

batch = [
    {
        'labels': torch.randn(4, 3),
        'index': 0
    },
    {
        'labels': torch.randn(4, 3),
        'index': 1
    }
]

ARGS = {
    'batch': batch
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
patchify Ran
ran on a synthesized input
microsoft/art-msra/multi_layer_gen/custom_model_transp_vae.py
code served (permissive licence) · get_code("91a80a2215fe15e4")
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, 192, 2, 2) 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 einops

torch.manual_seed(0)

# Test case 1: 4D tensor (batch, channels, height, width)
x_4d = torch.randn(2, 3, 16, 16)

# Test case 2: 3D tensor (channels, height, width)
x_3d = torch.randn(4, 24, 24)

ARGS = {
    "x": x_4d,
    "patch_size": 8
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, 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, 192, 2, 2) — shape (2, 192, 2, 2) 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
preprocess_Input Ran
ran on a synthesized input
microsoft/art-msra/layout_planner/inference_layout.py
code served (permissive licence) · get_code("d6905e2f00e0a901")
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 str. 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 re

def preprocess_Input(intention: str):
    intention = intention.replace('\n', '').replace('\r', '').replace('\\', '')
    intention = re.sub('\\.\\s*', '. ', intention)
    return intention

ARGS = {
    'intention': 'Hello world.\n This is a test\r with backslash \\ and dots.  Multiple spaces.'
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
unpatchify Ran
ran on a synthesized input
microsoft/art-msra/multi_layer_gen/custom_model_transp_vae.py
code served (permissive licence) · get_code("74720e4c7d36e702")
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, 32, 24) 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 einops

torch.manual_seed(0)

# Test case 1: 4D input (batch, channels*patch_size*patch_size, height, width)
x_4d = torch.randn(2, 64, 4, 3)  # bs=2, c*p1*p2=64 (8*8), h=4, w=3

# Test case 2: 3D input (channels*patch_size*patch_size, height, width)
x_3d = torch.randn(64, 4, 3)  # c*p1*p2=64 (8*8), h=4, w=3

ARGS = {
    "x": x_4d,
    "patch_size": 8,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, 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, 32, 24) — shape (2, 1, 32, 24) 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
get_dataset Not yet run
unverified (harvested, no recorded run)
microsoft/art-msra/layout_planner/training/datasets/layoutnew_dataset.py
code served (permissive licence) · get_code("7a4f7b082e40ac6b")
How it ran

no invocation recorded.

Also present in other papers' code

Of 14 samples lifted from this paper's linked repositories, 14 carry a function identity and 0 do not. Of the 14 distinct functions with an identity, 2 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 0 the same month, 0 later) and 12 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
l2_loss 2 other papers
2 dated earlier
2312.02252 (2023-12)
StoryGPT-V: Large Language Models as Consistent Story Visualizers
2404.18202 (2024-04)
WorldGPT: Empowering LLM as Multimodal World Model
accuracy 1 other paper
1 dated earlier
2402.18695 (2024-02)
Grounding Language Models for Visual Entity Recognition

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

Multi-layer image generation is a fundamental task that enables users to isolate, select, and edit specific image layers, thereby revolutionizing interactions with generative models. In this paper, we introduce the Anonymous Region Transformer (ART), which facilitates the direct generation of variable multi-layer transparent images based on a global text prompt and an anonymous region layout. Inspired by Schema theory 1 , this anonymous region layout allows the generative model to autonomously determine which set of visual tokens should align with which text tokens, which is in contrast to the previously dominant semantic layout for the image generation task. In addition, the layer-wise region crop mechanism, which only selects the visual tokens belonging to each anonymous region, significantly reduces attention computation costs and enables the efficient generation of images with numerous distinct layers (e.g., 50+). When compared to the full attention approach, our method is over 12 times faster and exhibits fewer layer conflicts. Furthermore, we propose a high-quality multi-layer transparent image autoencoder that supports the direct encoding and decoding of the transparency of variable multi-layer images in a joint manner. By enabling precise control and scalable layer generation, ART establishes a new paradigm for interactive content creation.

For agents

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

get_harvested_code_for_paper("2502.18364")
get_code_for_paper("2502.18364")
have("2502.18364")

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

curl https://syntology.ai/api/ran/2502.18364.json

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

[![Syntology run record](https://syntology.ai/api/ran/2502.18364.svg)](https://syntology.ai/paper/2502.18364)

Connect an agent — have() is free.