SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2503.09595 · 2025

PISA Experiments: Exploring Physics Post-Training for Video Diffusion Models by Watching Stuff Drop

arXiv · PDF · Open in the Atlas

Code that ran

8 of 8 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 4 have no contract verdict; 4 failures of our instrument; 0 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/2503.09595.json · badge: /api/ran/2503.09595.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
vision-x-nyu/pisa-experiments canonical (classifier label)
classifier basis not recorded on the link
8 ran · 0 no recorded run
FunctionStatusWhere it lives
binary_mask_IOU Ran
ran on a synthesized input
vision-x-nyu/pisa-experiments/utils/metrics.py
code served (permissive licence) · get_code("0008d9cdcc5af39e")
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 float64, ndim 0, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Construct valid arguments for binary_mask_IOU
# The function expects mask1 and mask2 with shape (f, n, height, width)
# where f is frames, n is number of masks, and height/width are 2D spatial dims

# Create small test masks with shape (2, 3, 4, 5)
# f=2 frames, n=3 masks, 4x5 spatial dimensions
f, n, h, w = 2, 3, 4, 5

# Create mask1: binary masks with some 1s and 0s
mask1 = np.zeros((f, n, h, w), dtype=np.uint8)
mask1[0, 0, 1:3, 1:3] = 1  # some overlap region
mask1[0, 1, 0:2, 0:2] = 1
mask1[0, 2, 2:4, 2:4] = 1
mask1[1, 0, 1:3, 1:4] = 1
mask1[1, 1, 0:3, 1:3] = 1
mask1[1, 2, 2:4, 0:2] = 1

# Create mask2: binary masks with some overlap and some different regions
mask2 = np.zeros((f, n, h, w), dtype=np.uint8)
mask2[0, 0, 1:3, 1:3] = 1  # exact match with mask1[0, 0]
mask2[0, 1, 0:2, 1:3] = 1  # partial overlap
mask2[0, 2, 1:3, 2:4] = 1  # partial overlap
mask2[1, 0, 1:3, 1:4] = 1  # exact match with mask1[1, 0]
mask2[1, 1, 0:2, 1:3] = 1  # partial overlap
mask2[1, 2, 3:4, 0:2] = 1  # partial overlap

ARGS = {
    'mask1': mask1,
    'mask2': mask2,
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
build_module Ran
ran; our contract draft was wrong, not the code
vision-x-nyu/pisa-experiments/opensora/registry.py
code served (permissive licence) · get_code("af810cd485700e94")
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 SimpleModule. 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
from copy import deepcopy


# Create a simple stand-in Registry class
class SimpleRegistry:
    def __init__(self):
        self.registry = {}
    
    def register(self, name):
        def decorator(cls):
            self.registry[name] = cls
            return cls
        return decorator
    
    def build(self, cfg):
        cfg = deepcopy(cfg)
        module_type = cfg.pop('type')
        if module_type in self.registry:
            return self.registry[module_type](**cfg)
        raise ValueError(f"Unknown module type: {module_type}")


# Create a simple test module
class SimpleModule(nn.Module):
    def __init__(self, in_features=4, out_features=3):
        super().__init__()
        self.linear = nn.Linear(in_features, out_features)
    
    def forward(self, x):
        return self.linear(x)


# Register the module
registry = SimpleRegistry()
registry.registry['SimpleModule'] = SimpleModule

# Create test instances for different scenarios
torch.manual_seed(0)

# Case 1: Module as dict config
module_dict = {
    'type': 'SimpleModule',
    'in_features': 4,
    'out_features': 3
}

# Case 2: Module as nn.Module instance
module_instance = SimpleModule(in_features=4, out_features=3)

# Define ARGS - test with dict input case
ARGS = {
    'module': module_dict,
    'builder': registry
}
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 SimpleModule — a SimpleModule output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a SimpleModule 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
dice_loss Ran
ran; our contract draft was wrong, not the code
vision-x-nyu/pisa-experiments/sam2/training/loss_fns.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from juneyeeHu/LM-EEC, is pointer only (licence: NONE) · get_code("e473e66fedc883c7")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from juneyeeHu/LM-EEC, 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.

iou_loss Ran
ran; our contract draft was wrong, not the code
vision-x-nyu/pisa-experiments/sam2/training/loss_fns.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from juneyeeHu/LM-EEC, is pointer only (licence: NONE) · get_code("b8f501ca29e77f1a")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from juneyeeHu/LM-EEC, 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.

mask_to_points Ran
ran on a synthesized input
vision-x-nyu/pisa-experiments/utils/metrics.py
code served (permissive licence) · get_code("67ba74e6741bc072")
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

# Create a small boolean mask
# The function uses np.where(mask == True) to find coordinates
# and stacks them into a (N, 2) array of points
mask = np.array([
    [True, False, True],
    [False, True, False],
    [True, False, False]
], dtype=bool)

ARGS = {
    "mask": mask
}
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
save_video_from_frames Ran
ran on a synthesized input
vision-x-nyu/pisa-experiments/data_processing/process_results.py
code served (permissive licence) · get_code("17da8dff24475bcc")
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 os
import tempfile
from PIL import Image
import numpy as np

# Create a temporary directory with dummy frame images
temp_dir = tempfile.mkdtemp()

# Generate a few dummy frames as JPEG files
for i in range(5):
    # Create a small random image
    img_array = np.random.randint(0, 255, (100, 100, 3), dtype=np.uint8)
    img = Image.fromarray(img_array)
    frame_path = os.path.join(temp_dir, f'{i:05d}.jpg')
    img.save(frame_path)

ARGS = {
    'frame_dir': temp_dir,
    'fps': 16
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy — 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
scaled_l2_distance Ran
ran on a synthesized input
vision-x-nyu/pisa-experiments/utils/metrics.py
code served (permissive licence) · get_code("7556f7e8aa8c3bc4")
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 float64, ndim 0, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# The function uses:
# - centroid1: a numpy array that supports subtraction and arithmetic operations
# - centroid2: a numpy array that supports subtraction and arithmetic operations
# - resolution: a scalar value used in division
# - scale_factor: a scalar value (has default)

# Create small distinct-sized numpy arrays for centroids
centroid1 = np.array([1.0, 2.0, 3.0, 4.0])
centroid2 = np.array([1.5, 2.5, 3.5, 4.5])

# resolution is a scalar
resolution = 0.5

# scale_factor has a default but we can provide one
scale_factor = 1 / np.sqrt(2)

ARGS = {
    "centroid1": centroid1,
    "centroid2": centroid2,
    "resolution": resolution,
    "scale_factor": scale_factor,
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
sigmoid_focal_loss Ran
ran; our contract draft was wrong, not the code
vision-x-nyu/pisa-experiments/sam2/training/loss_fns.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from juneyeeHu/LM-EEC, is pointer only (licence: NONE) · get_code("42fec6fdc1faf6f1")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from juneyeeHu/LM-EEC, 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.

Also present in other papers' code

Of 8 samples lifted from this paper's linked repositories, 8 carry a function identity and 0 do not. Of the 8 distinct functions with an identity, 4 are also present in other papers whose code we hold (4 with at least one paper dated earlier, 0 the same month, 3 later) and 4 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
dice_loss 5 other papers
4 dated earlier, 1 later
2302.01872 (2023-02)
MOSE: A New Dataset for Video Object Segmentation in Complex Scenes
2408.00714 (2024-08)
SAM 2: Segment Anything in Images and Videos
2408.07931 (2024-08)
Surgical SAM 2: Real-time Segment Anything in Surgical Video by Efficient Frame Pruning
and 2 more not named here
iou_loss 5 other papers
4 dated earlier, 1 later
2302.01872 (2023-02)
MOSE: A New Dataset for Video Object Segmentation in Complex Scenes
2408.00714 (2024-08)
SAM 2: Segment Anything in Images and Videos
2408.07931 (2024-08)
Surgical SAM 2: Real-time Segment Anything in Surgical Video by Efficient Frame Pruning
and 2 more not named here
sigmoid_focal_loss 5 other papers
4 dated earlier, 1 later
2302.01872 (2023-02)
MOSE: A New Dataset for Video Object Segmentation in Complex Scenes
2408.00714 (2024-08)
SAM 2: Segment Anything in Images and Videos
2408.07931 (2024-08)
Surgical SAM 2: Real-time Segment Anything in Surgical Video by Efficient Frame Pruning
and 2 more not named here
build_module 1 other paper
1 dated earlier
2410.08151 (2024-10)
Progressive Autoregressive Video Diffusion Models

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

Large-scale pre-trained video generation models excel in content creation but are not reliable as physically accurate world simulators out of the box. This work studies the process of post-training these models for accurate world modeling through the lens of the simple, yet fundamental, physics task of modeling object freefall. We show state-of-the-art video generation models struggle with this basic task, despite their visually impressive outputs. To remedy this problem, we find that fine-tuning on a relatively small amount of simulated videos is effective in inducing the dropping behavior in the model, and we can further improve results through a novel reward modeling procedure we introduce. Our study also reveals key limitations of post-training in generalization and distribution modeling. Additionally, we release a benchmark for this task that may serve as a useful diagnostic tool for tracking physical accuracy in large-scale video generative model development.

For agents

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

get_harvested_code_for_paper("2503.09595")
get_code_for_paper("2503.09595")
have("2503.09595")

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

curl https://syntology.ai/api/ran/2503.09595.json

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

[![Syntology run record](https://syntology.ai/api/ran/2503.09595.svg)](https://syntology.ai/paper/2503.09595)

Connect an agent — have() is free.