SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2109.02303 · ICCV · 2021

Encoder-decoder with Multi-level Attention for 3D Human Shape and Pose Estimation

Hongsheng Li, Shuai Yi, Zhengjia Li, Ziniu Wan, Maoqing Tian, Jianbo Liu

arXiv · PDF · Open in the Atlas

Code that ran

10 of 18 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 8 have no contract verdict; 2 failures of our instrument; 8 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/2109.02303.json · badge: /api/ran/2109.02303.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
ziniuwan/maed canonical (classifier label)
classifier basis: census
10 ran · 8 no recorded run
FunctionStatusWhere it lives
batch_adv_disc_l2_loss Ran
ran; our contract draft was wrong, not the code
ziniuwan/maed/lib/core/loss.py
code served (permissive licence) · get_code("cebcdf7656167a7a")
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 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)

# Create small tensors matching the expected shape N x 25
# real_disc_value: N x 25 where N=2
real_disc_value = torch.randn(2, 25)

# fake_disc_value: N x 25 where N=2
fake_disc_value = torch.randn(2, 25)

ARGS = {
    "real_disc_value": real_disc_value,
    "fake_disc_value": fake_disc_value,
}
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 1, 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
batch_encoder_disc_l2_loss Ran
ran; our contract draft was wrong, not the code
ziniuwan/maed/lib/core/loss.py
code served (permissive licence) · get_code("76d15c59d361a653")
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

torch.manual_seed(0)

# The function expects disc_value to be a tensor of shape N x 25
# where N is the batch size
# Let's create a small batch with N=2 and 25 dimensions
disc_value = torch.randn(2, 25)

ARGS = {
    "disc_value": disc_value
}
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
calc_kpt_bound Ran
ran on a synthesized input
ziniuwan/maed/lib/data_utils/penn_action_utils.py
code served (permissive licence) · get_code("6efb777aa4ff1f29")
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 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
import numpy as np

torch.manual_seed(0)

# Construct keypoint data: shape (num_keypoints, 3) where columns are [x, y, visibility]
# num_keypoints = 5, each has x, y, and visibility flag
kp_2d = torch.tensor([
    [100.0, 200.0, 1.0],    # visible keypoint
    [150.0, 250.0, 1.0],    # visible keypoint
    [120.0, 180.0, 0.0],    # invisible keypoint (will be skipped)
    [300.0, 350.0, 1.0],    # visible keypoint
    [50.0, 100.0, 1.0],     # visible keypoint
], dtype=torch.float32)

ARGS = {
    "kp_2d": kp_2d
}
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 tuple, ndim 1, 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
concat_mask Ran
ran on a synthesized input
ziniuwan/maed/lib/core/loss.py
code served (permissive licence) · get_code("feb703d44821fb72")
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)

# Create two dictionaries with matching keys
# masks_2d should contain 2D tensors (batch_size=2, features=4)
# masks_3d should contain 2D tensors (batch_size=2, features=4) that will be concatenated
masks_2d = {
    "mask_a": torch.randn(2, 4),
    "mask_b": torch.randn(2, 4),
    "mask_c": torch.randn(2, 4),
}

masks_3d = {
    "mask_a": torch.randn(2, 4),
    "mask_b": torch.randn(2, 4),
    "mask_c": torch.randn(2, 4),
}

ARGS = {
    "masks_2d": masks_2d,
    "masks_3d": masks_3d,
}
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
convert_kps_to_mask Ran
ran on a synthesized input
ziniuwan/maed/lib/data_utils/kp_utils.py
code served (permissive licence) · get_code("b4f1af8f64e4c25f")
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 (28, 28) 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

# Construct valid arguments for convert_kps_to_mask
kp_2d = np.array([
    [10, 20],    # keypoint 1
    [50, 100],   # keypoint 2
    [150, 180],  # keypoint 3
    [200, 210],  # keypoint 4
    [5, 5],      # keypoint 5
    [220, 220],  # keypoint 6 (near boundary)
    [0, 0],      # keypoint 7 (at boundary)
    [100, 100],  # keypoint 8
] + [[112, 112]] * 41, dtype=np.float32)  # pad to 49 keypoints

visibility = np.array([1, 1, 1, 1, 0, 1, 1, 1] + [1] * 41, dtype=np.int32)

mask_size = 28

ARGS = {
    "kp_2d": kp_2d,
    "visibility": visibility,
    "mask_size": mask_size,
}
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 ndarray, ndim 2, dtype family float, shape (28, 28) — shape (28, 28) 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_bbox_from_kp2d Ran
ran on a synthesized input
ziniuwan/maed/lib/data_utils/img_utils.py
code served (permissive licence) · get_code("ac0501b71e5ef134")
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 numpy as np

# Test case 1: 3D keypoints (batch of keypoint sets)
kp_2d_batch = np.random.rand(2, 4, 2) * 100  # batch_size=2, num_keypoints=4, coordinates=2

ARGS = {
    "kp_2d": kp_2d_batch
}
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 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
get_img_name Ran
ran on a synthesized input
ziniuwan/maed/lib/data_utils/insta_utils.py
code served (permissive licence) · get_code("b21bf2eb84721e30")
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 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 numpy as np
import os.path as osp

# Analyzing the function:
# - vid_name: a string that contains 'insta_variety/' which gets replaced
# - frame_id: an iterable (likely list/array) of frame identifiers
# The function constructs image paths by:
# 1. Replacing 'insta_variety/' with 'insta_variety_img/'
# 2. Splitting by '-' and reconstructing the path
# 3. Creating image filenames from frame_id values

ARGS = {
    'vid_name': 'insta_variety/video-001-clip',
    'frame_id': [0, 1, 2, 5]
}
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 ndarray, 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
load_mat Ran
ran on a synthesized input
ziniuwan/maed/lib/data_utils/penn_action_utils.py
code served (permissive licence) · get_code("8782aa2b6f3ea9ce")
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 tempfile
import scipy.io
import numpy as np

# Create a temporary .mat file with the expected structure
temp_file = tempfile.NamedTemporaryFile(suffix='.mat', delete=False)
temp_path = temp_file.name
temp_file.close()

# Create a minimal valid .mat file structure
mat_data = {
    'pose': np.array([[1.0, 2.0]]),
    '__header__': b'MATLAB 5.0 MAT-file',
    '__globals__': [],
    '__version__': '1.0',
    'train': np.array([1, 2, 3]),
    'action': np.array(['walk']),
    'nframes': np.array([[42]]),  # nframes is nested in the mat file
    'data': np.array([[1, 2, 3], [4, 5, 6]]),
}

scipy.io.savemat(temp_path, mat_data)

ARGS = {
    'path': temp_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — 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
pad_image Ran
ran on a synthesized input
ziniuwan/maed/lib/data_utils/img_utils.py
code served (permissive licence) · get_code("6578f5f4135e25e7")
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 (8, 6, 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 numpy as np

# Create a small test image with shape (height, width, channels)
# Using distinct dimensions: 4 height, 3 width, 2 channels
img = np.random.randint(0, 256, size=(4, 3, 2), dtype=np.uint8)

# Target padding dimensions - distinct sizes for probing
h = 8  # target height
w = 6  # target width

ARGS = {
    "img": img,
    "h": h,
    "w": w,
}
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 ndarray, ndim 3, dtype family int, shape (8, 6, 2) — shape (8, 6, 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
split_into_chunks Ran
ran on a synthesized input
ziniuwan/maed/lib/data_utils/img_utils.py
code served (permissive licence) · get_code("a5af3953b1583f4d")
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 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 numpy as np

# Analyzing the function:
# - vid_names: numpy array of video names (used with np.unique, indexing)
# - seqlen: integer for sequence length (used in arithmetic operations)
# - stride: integer for stride (used in view_as_windows)
# - pad: boolean (default True, used in if statement)

# Create sample data
# vid_names should be a numpy array of video identifiers
# Let's create an array with 8 elements, some repeated to test grouping
vid_names = np.array(['video_1', 'video_1', 'video_2', 'video_2', 
                      'video_2', 'video_3', 'video_3', 'video_3'])

# seqlen: sequence length for chunking
seqlen = 3

# stride: step size for sliding window
stride = 2

# pad: whether to pad sequences (default True)
pad = True

ARGS = {
    'vid_names': vid_names,
    'seqlen': seqlen,
    'stride': stride,
    'pad': pad
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, skimage — 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 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
keypoint_2d_hflip Not yet run
unverified (harvested, no recorded run)
ziniuwan/maed/lib/data_utils/kp_utils.py
code served (permissive licence) · get_code("100745c3a1bc859c")
How it ran

no invocation recorded.

keypoint_3d_hflip Not yet run
unverified (harvested, no recorded run)
ziniuwan/maed/lib/data_utils/kp_utils.py
code served (permissive licence) · get_code("9df8f844f3c2fae1")
How it ran

no invocation recorded.

read_calibration Not yet run
unverified (harvested, no recorded run)
ziniuwan/maed/lib/data_utils/mpii3d_utils.py
code served (permissive licence) · get_code("56adf5c78e74d698")
How it ran

no invocation recorded.

read_data Not yet run
unverified (harvested, no recorded run)
ziniuwan/maed/lib/data_utils/penn_action_utils.py
code served (permissive licence) · get_code("99675fb41689f0f0")
How it ran

no invocation recorded.

read_from_example Not yet run
unverified (harvested, no recorded run)
ziniuwan/maed/lib/data_utils/insta_utils.py
code served (permissive licence) · get_code("e8e628d7220b85a2")
How it ran

no invocation recorded.

read_openpose Not yet run
unverified (harvested, no recorded run)
ziniuwan/maed/lib/data_utils/mpii3d_utils.py
code served (permissive licence) · get_code("6af6cb56d2933f3e")
How it ran

no invocation recorded.

read_single_record Not yet run
unverified (harvested, no recorded run)
ziniuwan/maed/lib/data_utils/insta_utils.py
code served (permissive licence) · get_code("f9d14d3d4b62a24b")
How it ran

no invocation recorded.

update_cfg Not yet run
unverified (harvested, no recorded run)
ziniuwan/maed/lib/core/config.py
code served (permissive licence) · get_code("249e701b851ef72d")
How it ran

no invocation recorded.

Also present in other papers' code

Of 18 samples lifted from this paper's linked repositories, 18 carry a function identity and 0 do not. Of the 18 distinct functions with an identity, 5 are also present in other papers whose code we hold (3 with at least one paper dated earlier, 0 the same month, 2 later) and 13 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
update_cfg 4 other papers
4 later
2112.13715 (2021-12)
SmoothNet: A Plug-and-Play Network for Refining Human Poses in Videos
2203.08713 (2022-03)
DeciWatch: A Simple Baseline for 10x Efficient 2D and 3D Pose Estimation
and 1 more not named here
batch_adv_disc_l2_loss 1 other paper
1 dated earlier
2105.02467 (2021-05)
Body Meshes as Points
batch_encoder_disc_l2_loss 1 other paper
1 dated earlier
2105.02467 (2021-05)
Body Meshes as Points
read_calibration 1 other paper
1 later
2306.02850 (2023-06)
TRACE: 5D Temporal Regression of Avatars with Dynamic Cameras in 3D Environments
read_from_example 1 other paper
1 dated earlier
1812.01601 (2018-12)
Learning 3D Human Dynamics from Video

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

3D human shape and pose estimation is the essential task for human motion analysis, which is widely used in many 3D applications. However, existing methods cannot simultaneously capture the relations at multiple levels, including spatial-temporal level and human joint level. Therefore they fail to make accurate predictions in some hard scenarios when there is cluttered background, occlusion, or extreme pose. To this end, we propose Multi-level Attention Encoder-Decoder Network (MAED), including a Spatial-Temporal Encoder (STE) and a Kinematic Topology Decoder (KTD) to model multi-level attentions in a unified framework. STE consists of a series of cascaded blocks based on Multi-Head Self-Attention, and each block uses two parallel branches to learn spatial and temporal attention respectively. Meanwhile, KTD aims at modeling the joint level attention. It regards pose estimation as a top-down hierarchical process similar to SMPL kinematic tree. With the training set of 3DPW, MAED outperforms previous state-of-the-art methods by 6.2, 7.2, and 2.4 mm of PA-MPJPE on the three widely used benchmarks 3DPW, MPI-INF-3DHP, and Human3.6M respectively. Our code is available at https://github.com/ ziniuwan/maed.

For agents

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

get_harvested_code_for_paper("2109.02303")
get_code_for_paper("2109.02303")
have("2109.02303")

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

curl https://syntology.ai/api/ran/2109.02303.json

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

[![Syntology run record](https://syntology.ai/api/ran/2109.02303.svg)](https://syntology.ai/paper/2109.02303)

Connect an agent — have() is free.