gorilla-lab-scut / ttac Goto Github PK

View Code? Open in Web Editor NEW

39.0 1.0 3.0 1.35 MB

[NeurIPS 2022] Revisiting Realistic Test-Time Training: Sequential Inference and Adaptation by Anchored Clustering

Home Page: https://arxiv.org/abs/2206.02721

License: MIT License

Python 95.61% Shell 4.39%

domain-adaptation pseudo-labeling test-time-training deep-learning test-time-adaptation

ttac's Introduction

TTAC

This repository is an official implementation for our NeurIPS 2022 paper [Arxiv] [Openreview].

We implement a plug and play version of TTAC without queue on another work repo.

Revisiting Realistic Test-Time Training: Sequential Inference and Adaptation by Anchored Clustering

Yongyi Su¹ Xun Xu²¹ Kui Jia¹³
¹South China University of Technology ²Institute for Infocomm Research ³Peng Cheng Laboratory

Overview

CIFAR10/100

The code is released in the cifar folder.

ImageNet-C

The code is released in the imagenet folder.

Citation

If you find our work useful in your research, please consider citing:

@inproceedings{
  su2022revisiting,
  title={Revisiting Realistic Test-Time Training: Sequential Inference and Adaptation by Anchored Clustering},
  author={Yongyi Su and Xun Xu and Kui Jia},
  booktitle={Advances in Neural Information Processing Systems},
  editor={Alice H. Oh and Alekh Agarwal and Danielle Belgrave and Kyunghyun Cho},
  year={2022},
  url={https://openreview.net/forum?id=W-_4hgRkwb}
}

ttac's People

Contributors

Stargazers

Watchers

Forkers

nyquixt bostoncake better-chao

ttac's Issues

ModuleNotFoundError: No module named 'rich'

I installed your gorilla-core module and it worked well. However, when constructing the ImageNet dataset for training, when dealing with the frost corruption, I received following exception:

[ WARN:[email protected]] imread_('/home/username/miniconda3/envs/torch1121/lib/python3.7/site-packages/imagenet_c/frost/frost3.png'): can't open/read file: check file path/integrity
[ WARN:[email protected]] imread_('/home/username/miniconda3/envs/torch1121/lib/python3.7/site-packages/imagenet_c/frost/frost1.png'): can't open/read file: check file path/integrity[ WARN:[email protected]] imread_('/home/username/miniconda3/envs/torch1121/lib/python3.7/site-packages/imagenet_c/frost/frost2.png'): can't open/read file: check file path/integrity
Traceback (most recent call last):
File "/home/username/miniconda3/envs/torch1121/lib/python3.7/site-packages/gorilla/utils/processbar.py", line 198, in track
from rich.progress import track
ModuleNotFoundError: No module named 'rich'

I wonder how I could get the module rich, thanks!
By the way, when constructing the corruption datasets, I also received warnings as:

libpng warning: iCCP: profile 'icc': 'wtpt': ICC profile tag start not a multiple of 4
libpng warning: iCCP: profile 'icc': 'bkpt': ICC profile tag start not a multiple of 4
libpng warning: iCCP: profile 'icc': 'rXYZ': ICC profile tag start not a multiple of 4
libpng warning: iCCP: profile 'icc': 'gXYZ': ICC profile tag start not a multiple of 4
libpng warning: iCCP: profile 'icc': 'bXYZ': ICC profile tag start not a multiple of 4libpng warning: iCCP: profile 'icc': 'dmnd': ICC profile tag start not a multiple of 4
libpng warning: iCCP: profile 'icc': 'dmdd': ICC profile tag start not a multiple of 4
libpng warning: iCCP: profile 'icc': 'vued': ICC profile tag start not a multiple of 4libpng warning: iCCP: profile 'icc': 'view': ICC profile tag start not a multiple of 4
libpng warning: iCCP: profile 'icc': 'lumi': ICC profile tag start not a multiple of 4
libpng warning: iCCP: profile 'icc': 'meas': ICC profile tag start not a multiple of 4libpng warning: iCCP: profile 'icc': 'tech': ICC profile tag start not a multiple of 4
libpng warning: iCCP: profile 'icc': 'rTRC': ICC profile tag start not a multiple of 4
libpng warning: iCCP: profile 'icc': 'gTRC': ICC profile tag start not a multiple of 4libpng warning: iCCP: profile 'icc': 'bTRC': ICC profile tag start not a multiple of 4
libpng warning: iCCP: profile 'icc': 0h: PCS illuminant is not D50

I wonder if it's normal? Thanks!

Results Reproducibility

Hello,

Thank you for sharing your work.

I'm trying to reproduce ImageNet-C results. I downloaded ImageNet-C from here, and created a dataset using torchvision.datasets.ImageFolder and modified the __getitem__ to allow for is_carry_index.

First, evaluating pretrained ResNet50 (torchvision weights ImageNet1K-V1) on ImageNet-C (level 5) reports 83.11% error rate which is higher than the 82.22% reported in this repo. The error rate on the snow corruption using run_ttac_no_without_queue.sh is around 50% instead of 46.64%. The seed is fixed everywhere in the code so I'm expecting to reproduce both numbers. Any insights as to what might be wrong? I'm using PyTorch 1.12.1 and CUDA 10.2.89.

I reproduced the results on CIFAR-10 successfully.

Your suggestions and insights would be really appreciated.

Best regards,
Hani

AttributeError: module 'gorilla' has no attribute 'track'

When preparing the ImageNet data as instructed by the readme.md inside the imagenet/ folder, I receive the error AttributeError: module 'gorilla' has no attribute 'track'. I downloaded the gorilla==0.4.0 package from https://github.com/christophercrouzet/gorilla and it seems that it doesn't have an attribute named track. Would you please provide the gorilla package you are using? Thanks!

gorilla-lab-scut / ttac Goto Github PK

ttac's Introduction

TTAC

Revisiting Realistic Test-Time Training: Sequential Inference and Adaptation by Anchored Clustering

Overview

CIFAR10/100

ImageNet-C

Citation

ttac's People

Contributors

Stargazers

Watchers

Forkers

ttac's Issues

ModuleNotFoundError: No module named 'rich'

Results Reproducibility

AttributeError: module 'gorilla' has no attribute 'track'

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent