SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2408.12569 · ECCV · 2024

Sapiens: Foundation for Human Vision Models

Timur Bagautdinov, Stuart Anderson, Shunsuke Saito, Rawal Khirodkar, Julieta Martinez, Su Zhaoen, Austin James, Peter Selednik

arXiv · PDF · Open in the Atlas

Code that ran

9 of 11 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; 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/2408.12569.json · badge: /api/ran/2408.12569.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
ibaiGorordo/Sapiens-Pytorch-Inference community (classifier label)
classifier basis not recorded on the link
9 ran · 2 no recorded run
FunctionStatusWhere it lives
create_preprocessor Ran
ran on a synthesized input
ibaiGorordo/Sapiens-Pytorch-Inference/sapiens_inference/common.py
code served (permissive licence) · get_code("c95b90559e24d797")
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 Compose. 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 numpy as np
from torchvision import transforms

torch.manual_seed(0)

ARGS = {
    "input_size": (224, 224),
    "mean": [0.485, 0.456, 0.406],
    "std": [0.229, 0.224, 0.225],
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Compose — a Compose output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Compose 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
draw_depth_map Ran
ran on a synthesized input
ibaiGorordo/Sapiens-Pytorch-Inference/sapiens_inference/depth.py
code served (permissive licence) · get_code("f8b5db3cb31556d8")
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 (4, 5, 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 cv2
import numpy as np

# Create a sample depth map with realistic values
# The function expects an ndarray with depth values where 0 is treated specially
np.random.seed(0)
depth_map = np.random.rand(4, 5).astype(np.float32) * 100  # Random depths between 0-100
# Add some zeros to test the special case
depth_map[0, 0] = 0
depth_map[1, 1] = 0

ARGS = {
    "depth_map": depth_map
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, 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 (4, 5, 3) — shape (4, 5, 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
draw_normal_map Ran
ran on a synthesized input
ibaiGorordo/Sapiens-Pytorch-Inference/sapiens_inference/normal.py
code served (permissive licence) · get_code("ec008328da6ade0d")
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 (4, 3, 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 cv2
import numpy as np
import torch

torch.manual_seed(0)

# Create a small valid normal map
# Normal maps are typically 3-channel (RGB) with values in range [-1, 1]
# Shape should be (height, width, 3) for a single normal map
# Using small dimensions for testing
height, width = 4, 3
normal_map = np.random.randn(height, width, 3).astype(np.float32)

ARGS = {
    "normal_map": normal_map
}
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 (4, 3, 3) — shape (4, 3, 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
expand_boxes Ran
ran on a synthesized input
ibaiGorordo/Sapiens-Pytorch-Inference/sapiens_inference/predictor.py
code served (permissive licence) · get_code("e2ee7c364e6d3b69")
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 int, 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 numpy as np

# Construct boxes as a small numpy array with 2 boxes, each with 4 coordinates (x1, y1, x2, y2)
boxes = np.array([
    [10, 20, 100, 150],
    [50, 60, 200, 250]
], dtype=np.int32)

# Image shape as (height, width)
img_shape = (480, 640)

# Padding value
padding = 50

ARGS = {
    "boxes": boxes,
    "img_shape": img_shape,
    "padding": padding
}
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 int, 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
filter_small_boxes Ran
ran on a synthesized input
ibaiGorordo/Sapiens-Pytorch-Inference/sapiens_inference/predictor.py
code served (permissive licence) · get_code("904ba61aa2f9ac03")
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 int, 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 numpy as np

# Create sample boxes in format [x1, y1, x2, y2]
# We'll create boxes with varying heights relative to image height
boxes = np.array([
    [10, 20, 30, 100],    # height = 80
    [15, 50, 35, 60],     # height = 10 (small)
    [40, 10, 60, 150],    # height = 140
    [70, 80, 90, 85],     # height = 5 (very small)
])

img_height = 200
height_thres = 0.1

ARGS = {
    "boxes": boxes,
    "img_height": img_height,
    "height_thres": height_thres,
}
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 int, 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
pose_estimation_preprocessor Ran
ran on a synthesized input
ibaiGorordo/Sapiens-Pytorch-Inference/sapiens_inference/common.py
code served (permissive licence) · get_code("e95604d9afb269a8")
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 Compose. 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 numpy as np
from torchvision import transforms

torch.manual_seed(0)

ARGS = {
    "input_size": (224, 224),
    "mean": [0.485, 0.456, 0.406],
    "std": [0.229, 0.224, 0.225]
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Compose — a Compose output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Compose 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
postprocess_depth Ran
ran on a synthesized input
ibaiGorordo/Sapiens-Pytorch-Inference/sapiens_inference/depth.py
code served (permissive licence) · get_code("6de6c934cc5a5872")
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, 5) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np
import torch.nn.functional as F

torch.manual_seed(0)

# Create a small tensor to represent depth results
# The function expects results[0] to be indexable and have unsqueeze/squeeze methods
# It gets interpolated to img_shape, so it should have spatial dimensions
# Shape: (batch=1, channels=1, height=2, width=3) - small spatial dims
results = torch.randn(1, 1, 2, 3)

# img_shape is used as the target size for interpolation
img_shape = (4, 5)

ARGS = {
    "results": results,
    "img_shape": img_shape,
}
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, 5) — shape (4, 5) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
postprocess_normal Ran
ran on a synthesized input
ibaiGorordo/Sapiens-Pytorch-Inference/sapiens_inference/normal.py
code served (permissive licence) · get_code("b8a82028e4aade65")
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 float, shape (16, 20, 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
import numpy as np

torch.manual_seed(0)

# Create a tensor that matches what the function expects:
# - results[0] should be indexable and have .detach().cpu() methods
# - After squeeze(0), it should have shape (C, H, W) for interpolate
# - The function treats it as a batch of 1, so we need shape (1, C, H, W)
results = torch.randn(1, 3, 8, 12)

img_shape = (16, 20)

ARGS = {
    "results": results,
    "img_shape": img_shape
}
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 3, dtype family float, shape (16, 20, 3) — shape (16, 20, 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
postprocess_segmentation Ran
ran on a synthesized input
ibaiGorordo/Sapiens-Pytorch-Inference/sapiens_inference/segmentation.py
code served (permissive licence) · get_code("251244b8c5d68a56")
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 (7, 11) 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 torch.nn.functional as F

torch.manual_seed(0)

# The function expects:
# 1. results: torch.Tensor - will be indexed with [0], then unsqueezed to add batch dim
#    and interpolated. Should have shape like (batch, channels, height, width)
# 2. img_shape: tuple[int, int] - target (height, width) for interpolation

# Create a small results tensor with shape (batch=2, channels=4, height=3, width=5)
results = torch.randn(2, 4, 3, 5)

# Create a target image shape (different from source to test interpolation)
img_shape = (7, 11)

ARGS = {
    "results": results,
    "img_shape": img_shape,
}
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 (7, 11) — shape (7, 11) 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
download_hf_model Not yet run
unverified (harvested, no recorded run)
ibaiGorordo/Sapiens-Pytorch-Inference/sapiens_inference/common.py
code served (permissive licence) · get_code("95af9d8951a60dc5")
How it ran

no invocation recorded.

draw_boxes Not yet run
unverified (harvested, no recorded run)
ibaiGorordo/Sapiens-Pytorch-Inference/sapiens_inference/detector.py
code served (permissive licence) · get_code("43891ceb87a02a55")
How it ran

no invocation recorded.

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 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 11 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

We present Sapiens, a family of models for four fundamental human-centric vision tasks -2D pose estimation, body-part segmentation, depth estimation, and surface normal prediction. Our models natively support 1K high-resolution inference and are extremely easy to adapt for individual tasks by simply fine-tuning foundation models pretrained on over 300 million in-the-wild human images. We observe that, given the same computational budget, self-supervised pretraining on a curated dataset of human images significantly boosts the performance for a diverse set of human-centric tasks. The resulting models exhibit remarkable generalization to in-the-wild data, even when labeled data is scarce or entirely synthetic. Our simple model design also brings scalability -model performance across tasks significantly improves as we scale the number of parameters from 0.3 to 2 billion. Sapiens consistently surpasses existing complex baselines across various human-centric benchmarks. Specifically, we achieve significant improvements over the prior state-of-the-art on Humans-5K (pose) by 7.6 mAP, Humans-2K (part-seg) by 17.1 mIoU, Hi4D (depth) by 22.4% relative RMSE, and THuman2 (normal) by 53.5% relative angular error.

For agents

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

get_harvested_code_for_paper("2408.12569")
get_code_for_paper("2408.12569")
have("2408.12569")

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

curl https://syntology.ai/api/ran/2408.12569.json

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

[![Syntology run record](https://syntology.ai/api/ran/2408.12569.svg)](https://syntology.ai/paper/2408.12569)

Connect an agent — have() is free.