SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2302.10586 · NeurIPS · 2023

Diffusion Models and Semi-Supervised Learners Benefit Mutually with Few Labels

Jun Zhu, Chongxuan Li, Fan Bao, Jiacheng Sun, Zebin You, Yong Zhong

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 22 functions out of this paper's own repositories and ran 17 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
ML-GSAI/DPT canonical 11 of 16
copy not recorded — 3 of 3
baofff/u-vit extension 3 of 3
FunctionStatusWhere it lives
center_crop Ran ML-GSAI/DPT/datasets.py
code served (permissive licence) · get_code("f057bf3471d77c43")
distributed_sinkhorn Ran ML-GSAI/DPT/src/losses.py
code served (permissive licence) · get_code("7478f8e64cee8964")
drop_path Ran ML-GSAI/DPT/src/deit.py
code served (permissive licence) · get_code("55120f2026b56aa2")
get_sde Ran ML-GSAI/DPT/sde.py
code served (permissive licence) · get_code("b723a1cb4778012f")
interpolate_fn Ran ML-GSAI/DPT/dpm_solver_pp.py
code served (permissive licence) · get_code("86c7877e164664bc")
make_transforms Ran ML-GSAI/DPT/src/data_manager.py
code served (permissive licence) · get_code("cec1c5fdf881ddf4")
model_wrapper Ran ML-GSAI/DPT/dpm_solver_pp.py
code served (permissive licence) · get_code("ace7da590f4070e0")
model_wrapper Ran ML-GSAI/DPT/dpm_solver_pytorch.py
code served (permissive licence) · get_code("b102ed94bbf0cfb9")
mos Ran ML-GSAI/DPT/sde.py
code served (permissive licence) · get_code("02c82bdf3494ea71")
patchify Ran this paper's copy was not recorded; identical code first harvested from baofff/U-ViT
pointer only · get_code("7a7de2a8e080f047")
patchify Ran baofff/u-vit/libs/uvit.py
code served (permissive licence) · get_code("9cf076a2ae913f94")
random_crop_arr Ran ML-GSAI/DPT/datasets.py
code served (permissive licence) · get_code("05d1f95a391c0ec3")
stable_diffusion_beta_schedule Ran ML-GSAI/DPT/sample_ldm_discrete_all.py
code served (permissive licence) · get_code("2ee6e353f1dd4cf3")
timestep_embedding Ran this paper's copy was not recorded; identical code first harvested from baofff/U-ViT
pointer only · get_code("262b18d8278dcf70")
timestep_embedding Ran baofff/u-vit/libs/uvit.py
code served (permissive licence) · get_code("b494dd297ded5960")
unpatchify Ran this paper's copy was not recorded; identical code first harvested from baofff/U-ViT
pointer only · get_code("82e1ad8a9b00554e")
unpatchify Ran baofff/u-vit/libs/uvit.py
code served (permissive licence) · get_code("f9244c08c1011341")
center_crop_arr Not yet run ML-GSAI/DPT/datasets.py
code served (permissive licence) · get_code("f8b4a29a52612a41")
copy_imgnt_locally Not yet run ML-GSAI/DPT/src/data_manager.py
code served (permissive licence) · get_code("b2c367ab087ec039")
init_data Not yet run ML-GSAI/DPT/src/data_manager.py
code served (permissive licence) · get_code("ec08cc8c5589ddfe")
load_checkpoint Not yet run ML-GSAI/DPT/src/msn_train.py
code served (permissive licence) · get_code("e22a49db490231b6")
stp Not yet run ML-GSAI/DPT/sde.py
code served (permissive licence) · get_code("4d33c2f525f1b1c0")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

In an effort to further advance semi-supervised generative and classification tasks, we propose a simple yet effective training strategy called dual pseudo training (DPT), built upon strong semi-supervised learners and diffusion models. DPT operates in three stages: training a classifier on partially labeled data to predict pseudo-labels; training a conditional generative model using these pseudo-labels to generate pseudo images; and retraining the classifier with a mix of real and pseudo images. Empirically, DPT consistently achieves SOTA performance of semi-supervised generation and classification across various settings. In particular, with one or two labels per class, DPT achieves a Fréchet Inception Distance (FID) score of 3.08 or 2.52 on ImageNet 256 × 256. Besides, DPT outperforms competitive semi-supervised baselines substantially on ImageNet classification tasks, achieving top-1 accuracies of 59.0 (+2.8), 69.5 (+3.0), and 74.4 (+2.0) with one, two, or five labels per class, respectively. Notably, our results demonstrate that diffusion can generate realistic images with only a few labels (e.g., < 0.1%) and generative augmentation remains viable for semi-supervised classification. Our code is available at https://github.com/ML-GSAI/DPT.

For agents

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

get_harvested_code_for_paper("2302.10586")
get_code_for_paper("2302.10586")
have("2302.10586")

Connect an agent — have() is free.