SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2303.05938 · CVPR · 2023

ACR: Attention Collaboration-based Regressor for Arbitrary Two-Hand Reconstruction

Jue Wang, Shaoli Huang, Toby Breckon, Zhengdi Yu, Chen Fang

arXiv · PDF · Open in the Atlas

Code that ran

12 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 9 have no contract verdict; 3 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/2303.05938.json · badge: /api/ran/2303.05938.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
zhengdiyu/arbitrary-hands-3d-reconstruction canonical (classifier label)
classifier basis not recorded on the link
12 ran · 2 no recorded run
FunctionStatusWhere it lives
collect_image_list Ran
ran on a synthesized input
zhengdiyu/arbitrary-hands-3d-reconstruction/acr/utils.py
code served (permissive licence) · get_code("7df23e797e46d689")
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 os
import tempfile
import shutil

# Create a temporary directory structure with some image files
temp_dir = tempfile.mkdtemp()

# Create some subdirectories and files
subdir1 = os.path.join(temp_dir, 'subdir1')
subdir2 = os.path.join(temp_dir, 'subdir2')
os.makedirs(subdir1)
os.makedirs(subdir2)

# Create some dummy image files with valid extensions
image_files = [
    os.path.join(temp_dir, 'image1.jpg'),
    os.path.join(temp_dir, 'image2.png'),
    os.path.join(subdir1, 'image3.jpeg'),
    os.path.join(subdir2, 'image4.bmp'),
]

for img_file in image_files:
    with open(img_file, 'w') as f:
        f.write('dummy image data')

# Create some non-image files that should be ignored
non_image_files = [
    os.path.join(temp_dir, 'readme.txt'),
    os.path.join(subdir1, 'data.csv'),
]

for non_img_file in non_image_files:
    with open(non_img_file, 'w') as f:
        f.write('not an image')

ARGS = {
    'image_folder': temp_dir,
    'collect_subdirs': True,
}
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 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
conv3x3 Ran
ran; our contract draft was wrong, not the code
zhengdiyu/arbitrary-hands-3d-reconstruction/acr/model.py
pointer only (licence: MIT); the stored copy get_code serves, from lixiaotong97/DSU, is pointer only (licence: NONE) · get_code("fac5364e2f53c6db")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from lixiaotong97/DSU, and GitHub detected no licence in that repository when it was checked. 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.

gather_feature Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zhengdiyu/arbitrary-hands-3d-reconstruction/acr/result_parser.py
pointer only (licence: MIT); the stored copy get_code serves, from jwh97nn/DeepDPS, is pointer only (licence: NONE) · get_code("e7fc07c30373e482")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from jwh97nn/DeepDPS, and GitHub detected no licence in that repository when it was checked. 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_coord_maps Ran
ran on a synthesized input
zhengdiyu/arbitrary-hands-3d-reconstruction/acr/model.py
code served (permissive licence) · get_code("b77632bd876028e5")
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 (1, 2, 128, 128) 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)

ARGS = {
    "size": 128
}
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 (1, 2, 128, 128) — shape (1, 2, 128, 128) 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_keypoint_rgb Ran
ran on a synthesized input
zhengdiyu/arbitrary-hands-3d-reconstruction/acr/visualization.py
code served (permissive licence) · get_code("9f74c89bce4cb583")
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

# Construct a skeleton structure that the function expects
# The function iterates through skeleton, accessing skeleton[i]['name']
skeleton = [
    {'name': 'hand_thumb_null'},
    {'name': 'hand_thumb3'},
    {'name': 'hand_thumb2'},
    {'name': 'hand_thumb1'},
    {'name': 'hand_thumb0'},
    {'name': 'hand_index_null'},
    {'name': 'hand_index3'},
    {'name': 'hand_index2'},
    {'name': 'hand_index1'},
    {'name': 'hand_middle_null'},
    {'name': 'hand_middle3'},
    {'name': 'hand_middle2'},
    {'name': 'hand_middle1'},
    {'name': 'hand_ring_null'},
    {'name': 'hand_ring3'},
    {'name': 'hand_ring2'},
    {'name': 'hand_ring1'},
    {'name': 'hand_pinky_null'},
    {'name': 'hand_pinky3'},
    {'name': 'hand_pinky2'},
    {'name': 'hand_pinky1'},
    {'name': 'hand_wrist'},
]

ARGS = {
    'skeleton': skeleton
}
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
get_kp2d_on_org_img Ran
ran on a synthesized input
zhengdiyu/arbitrary-hands-3d-reconstruction/acr/utils.py
code served (permissive licence) · get_code("4fe8bffaef32f549")
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 ndarray, ndim 2, dtype family float, shape (4, 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 numpy as np

torch.manual_seed(0)

# Create kp2d: K x 2 keypoints (normalized coordinates in [-1, 1] range)
# Using K=4 keypoints with 2D coordinates
kp2d = np.array([
    [-0.5, 0.3],
    [0.2, -0.8],
    [0.7, 0.1],
    [-0.9, 0.6]
], dtype=np.float32)

# Create offset: a tensor/array with 11 elements as unpacked in the function
# pad_size_h, pad_size_w, lt_h, rb_h, lt_w, rb_w, offset_h, size_h, offset_w, size_w, length
offset = torch.tensor([
    256,      # pad_size_h
    512,      # pad_size_w
    10,       # lt_h
    246,      # rb_h
    20,       # lt_w
    492,      # rb_w
    5,        # offset_h
    236,      # size_h
    8,        # offset_w
    472,      # size_w
    100       # length
], dtype=torch.float32)

ARGS = {
    'kp2d': kp2d,
    'offset': offset
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (4, 2) — shape (4, 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
make_heatmaps Ran
ran on a synthesized input
zhengdiyu/arbitrary-hands-3d-reconstruction/acr/visualization.py
code served (permissive licence) · get_code("efde4255027e258d")
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 ndarray, ndim 3, dtype family int, shape (32, 24, 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 numpy as np
import torch
import cv2

torch.manual_seed(0)

# Create a small image (height=32, width=24, channels=3)
image = np.random.randint(0, 256, size=(32, 24, 3), dtype=np.uint8)

# Create heatmaps tensor (num_joints=4, small spatial dims=8x6)
heatmaps = torch.rand(4, 8, 6)

ARGS = {
    "image": image,
    "heatmaps": heatmaps,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family int, shape (32, 24, 3) — shape (32, 24, 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
make_tagmaps Ran
ran on a synthesized input
zhengdiyu/arbitrary-hands-3d-reconstruction/acr/visualization.py
code served (permissive licence) · get_code("e1957e87a34993e1")
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 ndarray, ndim 3, dtype family int, shape (32, 120, 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 numpy as np
import cv2
import torch

torch.manual_seed(0)

# Create a small test image (height=32, width=24, channels=3)
image = np.random.randint(0, 256, (32, 24, 3), dtype=np.uint8)

# Create tagmaps tensor (num_joints=4, height=32, width=24)
# Using torch tensor since the code calls .add(), .div(), .mul(), .clamp(), .byte(), .detach(), .cpu(), .numpy()
tagmaps = torch.randn(4, 32, 24, dtype=torch.float32)

ARGS = {
    "image": image,
    "tagmaps": tagmaps
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family int, shape (32, 120, 3) — shape (32, 120, 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
nms Ran
ran on a synthesized input
zhengdiyu/arbitrary-hands-3d-reconstruction/acr/result_parser.py
code served (permissive licence) · get_code("8bfe96e53c77a01b")
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, 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.functional as F

torch.manual_seed(0)

# Create a small detection tensor (batch_size=2, num_detections=4)
# Each detection has a score value for NMS
det = torch.tensor([
    [0.9, 0.7, 0.5, 0.3],
    [0.8, 0.6, 0.4, 0.2]
], dtype=torch.float32)

# pool_func should take a tensor and return max values
# Using max_pool1d as a pool function that finds max along dimension
def pool_func(x):
    # Reshape for max_pool1d: (batch, channels, length) -> add channel dim
    x_reshaped = x.unsqueeze(1)  # (2, 1, 4)
    pooled = F.max_pool1d(x_reshaped, kernel_size=x.shape[1], stride=1)
    # pooled shape: (2, 1, 1), squeeze to (2, 1)
    return pooled.squeeze(-1).expand_as(x)

ARGS = {
    "det": det,
    "pool_func": pool_func
}
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, 4) — shape (2, 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
normalize_kps Ran
ran on a synthesized input
zhengdiyu/arbitrary-hands-3d-reconstruction/acr/utils.py
code served (permissive licence) · get_code("bda21cbd22b37231")
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, 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 numpy as np

torch.manual_seed(0)

# Create a small keypoints tensor
# Shape: (batch_size, num_keypoints_with_confidence)
# kps[:, 0] and kps[:, 1] are x and y coordinates
# kps[:, 2] is confidence score (optional)
kps = torch.tensor([
    [100.0, 150.0, 0.9],
    [200.0, 250.0, 0.05]
], dtype=torch.float32)

# Image shape as a tuple (height, width)
image_shape = (480, 640)

# Resize parameter
resize = 512

# set_minus parameter
set_minus = True

ARGS = {
    "kps": kps,
    "image_shape": image_shape,
    "resize": resize,
    "set_minus": set_minus
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, 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, 2) — shape (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
th_pack Ran
ran; our contract draft was wrong, not the code
zhengdiyu/arbitrary-hands-3d-reconstruction/mano/manolayer.py
pointer only (licence: MIT); the stored copy get_code serves, from reyuwei/PIANO_model, is pointer only (licence: NONE) · get_code("a012fac19e3c7477")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from reyuwei/PIANO_model, and GitHub detected no licence in that repository when it was checked. 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.

th_posemap_axisang Ran
ran on a synthesized input
zhengdiyu/arbitrary-hands-3d-reconstruction/mano/manolayer.py
code served (permissive licence) · get_code("42521f4ec8660580")
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, ndim 3, 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)

# pose_vectors should be a tensor of shape [batch_size, num_joints * 3]
# where num_joints is the number of joints (axis-angle representation uses 3 values per joint)
# The function will reshape it to [-1, 3] for batch_rodrigues, then back to [batch_size, num_joints * 9]

batch_size = 2
num_joints = 4
pose_vectors = torch.randn(batch_size, num_joints * 3, dtype=torch.float32)

ARGS = {
    "pose_vectors": pose_vectors
}
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, ndim 3, dtype family float — a tuple 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
parse_args Not yet run
unverified (harvested, no recorded run)
zhengdiyu/arbitrary-hands-3d-reconstruction/acr/config.py
code served (permissive licence) · get_code("e65d2c00abafb2f5")
How it ran

no invocation recorded.

th_with_zeros Not yet run
unverified (harvested, no recorded run)
zhengdiyu/arbitrary-hands-3d-reconstruction/mano/manolayer.py
pointer only (licence: MIT); the stored copy get_code serves, from reyuwei/PIANO_model, is pointer only (licence: NONE) · get_code("cd37580add3ccba5")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from reyuwei/PIANO_model, and GitHub detected no licence in that repository when it was checked. 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 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, 5 are also present in other papers whose code we hold (4 with at least one paper dated earlier, 1 the same month, 3 later) and 9 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
conv3x3 686 other papers
459 dated earlier, 10 the same month, 210 later, 7 same year, order not determinable · 31 only as an assembly brick
1406.2661 (2014-06)
Generative Adversarial Networks
1409.0575 (2014-09)
ImageNet Large Scale Visual Recognition Challenge
1412.6572 (2014-12)
Explaining and Harnessing Adversarial Examples
and 683 more not named here
gather_feature 3 other papers
2 dated earlier, 1 later
2108.07682 (2021-08)
Fully Convolutional Networks for Panoptic Segmentation with Point-based Supervision
2206.00468 (2022-06)
PanopticDepth: A Unified Framework for Depth-aware Panoptic Segmentation
2307.14786 (2023-07)
Towards Deeply Unified Depth-aware Panoptic Segmentation with Bi-directional Guidance Learning
th_pack 2 other papers
2 dated earlier
2106.10893 (2021-06)
PIANO: A Parametric Hand Bone Model from Magnetic Resonance Imaging
2302.00988 (2023-02)
HaMuCo: Hand Pose Estimation via Multiview Collaborative Self-Supervised Learning
th_with_zeros 2 other papers
2 dated earlier
2106.10893 (2021-06)
PIANO: A Parametric Hand Bone Model from Magnetic Resonance Imaging
2302.00988 (2023-02)
HaMuCo: Hand Pose Estimation via Multiview Collaborative Self-Supervised Learning
get_coord_maps 1 other paper
1 later
2306.02850 (2023-06)
TRACE: 5D Temporal Regression of Avatars with Dynamic Cameras in 3D Environments

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

Reconstructing two hands from monocular RGB images is challenging due to frequent occlusion and mutual confusion. Existing methods mainly learn an entangled representation to encode two interacting hands, which are incredibly fragile to impaired interaction, such as truncated hands, separate hands, or external occlusion. This paper presents ACR (Attention Collaboration-based Regressor), which makes the first attempt to reconstruct hands in arbitrary scenarios. To achieve this, ACR explicitly mitigates interdependencies between hands and between parts by leveraging center and part-based attention for feature extraction. However, reducing interdependence helps release the input constraint while weakening the mutual reasoning about reconstructing the interacting hands. Thus, based on center attention, ACR also learns cross-hand prior that handle the interacting hands better. We evaluate our method on various types of hand reconstruction datasets. Our method significantly outperforms the best interacting-hand approaches on the InterHand2.6M dataset while yielding comparable performance with the state-ofthe-art single-hand methods on the FreiHand dataset. More qualitative results on in-the-wild and hand-object interaction datasets and web images/videos further demonstrate the effectiveness of our approach for arbitrary hand reconstruction. Our code is available at this link 1 .

For agents

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

get_harvested_code_for_paper("2303.05938")
get_code_for_paper("2303.05938")
have("2303.05938")

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

curl https://syntology.ai/api/ran/2303.05938.json

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

[![Syntology run record](https://syntology.ai/api/ran/2303.05938.svg)](https://syntology.ai/paper/2303.05938)

Connect an agent — have() is free.