Pytorch implementation of "Training a 85.4% Top-1 Accuracy Vision Transformer with 56M Parameters on ImageNet"

Overview

Token Labeling: Training an 85.4% Top-1 Accuracy Vision Transformer with 56M Parameters on ImageNet (arxiv)

This is a Pytorch implementation of our technical report.

Compare

Comparison between the proposed LV-ViT and other recent works based on transformers. Note that we only show models whose model sizes are under 100M.

Training Pipeline

Pipeline

Our codes are based on the pytorch-image-models by Ross Wightman.

LV-ViT Models

Model layer dim Image resolution Param Top 1 Download
LV-ViT-S 16 384 224 26.15M 83.3 link
LV-ViT-S 16 384 384 26.30M 84.4 link
LV-ViT-M 20 512 224 55.83M 84.0 link
LV-ViT-M 20 512 384 56.03M 85.4 link
LV-ViT-L 24 768 448 150.47M 86.2 link

Requirements

torch>=1.4.0 torchvision>=0.5.0 pyyaml timm==0.4.5

data prepare: ImageNet with the following folder structure, you can extract imagenet by this script.

│imagenet/
├──train/
│  ├── n01440764
│  │   ├── n01440764_10026.JPEG
│  │   ├── n01440764_10027.JPEG
│  │   ├── ......
│  ├── ......
├──val/
│  ├── n01440764
│  │   ├── ILSVRC2012_val_00000293.JPEG
│  │   ├── ILSVRC2012_val_00002138.JPEG
│  │   ├── ......
│  ├── ......

Validation

Replace DATA_DIR with your imagenet validation set path and MODEL_DIR with the checkpoint path

CUDA_VISIBLE_DEVICES=0 bash eval.sh /path/to/imagenet/val /path/to/checkpoint

Label data

We provide NFNet-F6 generated dense label map here. As NFNet-F6 are based on pure ImageNet data, no extra training data is involved.

Training

Coming soon

Reference

If you use this repo or find it useful, please consider citing:

@misc{jiang2021token,
      title={Token Labeling: Training an 85.4% Top-1 Accuracy Vision Transformer with 56M Parameters on ImageNet}, 
      author={Zihang Jiang and Qibin Hou and Li Yuan and Daquan Zhou and Xiaojie Jin and Anran Wang and Jiashi Feng},
      year={2021},
      eprint={2104.10858},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Related projects

T2T-ViT, Re-labeling ImageNet.

Comments
  • error: download the pretrained model but couldn't be unzipped

    error: download the pretrained model but couldn't be unzipped

    tar -xvf lvvit_s-26M-384-84-4.pth.tar tar: This does not look like a tar archive tar: Skipping to next header tar: Exiting with failure status due to previous errors

    opened by Williamlizl 10
  • The accuracy of the validation set is 0,and the loss is always around 13

    The accuracy of the validation set is 0,and the loss is always around 13

    Hello! I use ILSVRC2012_img_train and ILSVRC2012_img_val, and use the provided label_top5_train_nfnet from Google Drive. I train lv-vit-s with batch_size 64 without apex for one epoch. Thanks for your advice.

    opened by yifanQi98 7
  • Pretrained weights for LV-ViT-T

    Pretrained weights for LV-ViT-T

    Hi,

    Thanks for sharing your work. Could you also provide the pre-trained weights for the LV-ViT-T model variant, the one that achieves 79.1% top1-acc. as mentioned in Table 1 of your paper?

    All the best, Marc

    opened by marc345 5
  • train error: AttributeError: 'tuple' object has no attribute 'log_softmax'

    train error: AttributeError: 'tuple' object has no attribute 'log_softmax'

    Hi, thanks for you great work. When I train script, some error occurs: AttributeError: 'tuple' object has no attribute 'log_softmax'

    with amp_autocast():   
                output = model(input)  
                loss = loss_fn(output, target)  # error occurs
    
    

    and loss function is train_loss_fn = LabelSmoothingCrossEntropy(smoothing=0.0).cuda()

    by the way: Could you please tell me why we need to specify smoothing=0.0?

    opened by lxy5513 5
  • RuntimeError: CUDA error: device-side assert triggered

    RuntimeError: CUDA error: device-side assert triggered

    I am a green hand of DL. When I run the code of volo with tlt in a single or multi GPU, I get an error as follows: /pytorch/aten/src/ATen/native/cuda/ScatterGatherKernel.cu:312: operator(): block: [0,0,0], thread: [25,0,0] Assertion idx_dim >= 0 && idx_dim < index_size && "index out of bounds" failed. Traceback (most recent call last): File "main.py", line 949, in main() File "main.py", line 664, in main optimizers=optimizers) File "main.py", line 773, in train_one_epoch label_size=args.token_label_size) File "/opt/conda/lib/python3.6/site-packages/tlt/data/mixup.py", line 90, in mixup_target y1 = get_labelmaps_with_coords(target, num_classes, on_value=on_value, off_value=off_value, device=device, label_size=label_size) File "/opt/conda/lib/python3.6/site-packages/tlt/data/mixup.py", line 64, in get_labelmaps_with_coords num_classes=num_classes,device=device) File "/opt/conda/lib/python3.6/site-packages/tlt/data/mixup.py", line 16, in get_featuremaps _label_topk[1][:, :, :].long(), RuntimeError: CUDA error: device-side assert triggered.

    I can't fix this problem right now.

    opened by JIAOJIAYUASD 4
  • Generating label for custom dataset

    Generating label for custom dataset

    Hello,

    Thank you for sharing your work. I am currently trying to generate token label to a custom dataset for model lvvit_s, but I keep getting the loss close to 7 and the Accuracy 0 (not pre-trained and using 1 GPU in Google Colab). I also tried using the pre-trained model with --transfer but got 0 in both Loss and Acc . What option should I use for a custom dataset? image

    opened by AleMaiaF 2
  • generate_label.py unable to find model lvvit_s

    generate_label.py unable to find model lvvit_s

    Hi,

    When I tried to run the label generation script for the model lvvit_s it returned an error "RuntimeError: Unknown model".

    Solution: It worked when I added the line "import tlt.models" in the file generate_label.py.

    opened by AleMaiaF 2
  • Can Token labeling reach higher than annotator model?

    Can Token labeling reach higher than annotator model?

    Greetings,

    Thank you for this incredible research.

    I would like to know if it is possible to use Token Labeling to achieve scores higher than that of the annotator model, I believe this was the case with VOLO D5 model where it achieved higher score than NFNet, model used for annotation.

    opened by ErenBalatkan 1
  • label_map does not do the same augmentation (random crop) as the input image

    label_map does not do the same augmentation (random crop) as the input image

    Hi Thanks so much for the nice work! I am curious if you could share the insight on processing of the label_map. If I understand it correctly, after we load image and the corresponding, we shall do the same cropping/ flip/ resize, but in https://github.com/zihangJiang/TokenLabeling/blob/aa438eff9b9fc2daa8c8b4cc6bfaa6e3721f995e/tlt/data/label_transforms_factory.py#L58-L73 Seems only image was cropped, but the label map does not do the same cropping, which make the label map not match with the image?

    Shall we do

            return torchvision_F.resized_crop(
                    img, i, j, h, w, self.size, interpolation
            ), torchvision_F.resized_crop(
                    label_map, i / ratio, j / ratio, h / ratio, w / ratio, self.size, interpolation
            )
    

    Thanks

    opened by haooooooqi 1
  • Python3.6, ok; Python3.8, error

    Python3.6, ok; Python3.8, error

    Test: [ 0/1] Time: 11.293 (11.293) Loss: 0.7043 (0.7043) [email protected]: 42.1875 (42.1875) [email protected]: 100.0000 (100.0000) Test: [ 1/1] Time: 0.108 (5.701) Loss: 0.5847 (0.6689) [email protected]: 89.8148 (56.3187) [email protected]: 100.0000 (100.0000) free(): invalid pointer free(): invalid pointer Traceback (most recent call last): File "/opt/conda/lib/python3.8/runpy.py", line 194, in _run_module_as_main return _run_code(code, main_globals, None, File "/opt/conda/lib/python3.8/runpy.py", line 87, in _run_code exec(code, run_globals) File "/opt/conda/lib/python3.8/site-packages/torch/distributed/launch.py", line 303, in <module> main() File "/opt/conda/lib/python3.8/site-packages/torch/distributed/launch.py", line 294, in main raise subprocess.CalledProcessError(returncode=process.returncode, subprocess.CalledProcessError: Command '['/opt/conda/bin/python3.8', '-u', 'main.py', '--local_rank=1', './dataset/c/c', '--model', 'lvvit_s', '-b', '128', '--apex-amp', '--img-size', '224', '--drop-path', '0.1', '--token-label', '--token-label-size', '14', '--dense-weight', '0.0', '--num-classes', '2', '--finetune', './pretrained/lvvit_s-26M-384-84-4.pth.tar']' died with <Signals.SIGABRT: 6>. [email protected]:/puxin_libochao/TokenLabeling# CUDA_VISIBLE_DEVICES=0,1 bash ./distributed_train.sh 2 ./dataset/c/c --model lvvit_s -b 128 --apex-amp --img-size 224 --drop-path 0.1 --token-label --token-label-size 14 --dense-weight 0.0 --num-classes 2 --finetune ./pretrained/lvvit_s-26M-384-84-4.pth.tar

    opened by Williamlizl 1
  • A Bag of Training Techniques for ViT

    A Bag of Training Techniques for ViT

    Hi, thanks for your wonderful work. I have a question that whether training techniques mentioned in the LV-Vit can be used in other downstream task like object detection? In your paper, I see that many of this techniques are used in ImageNet. Thanks!

    opened by qdd1234 1
  • how to apply token labeling to CNN ?

    how to apply token labeling to CNN ?

    Hello ~ I'm interested in your token labeling technique, So I want to apply this technique in CNN based model because ViT is very heavy to train.

    can I get the your code with CNN token labeling? if you're not give me some detail for implementing

    thank you.

    opened by HoJ00n2 0
  • Model settings for Cifar10

    Model settings for Cifar10

    I am interested if there is any LV-ViT- model setup you have tested for Cifar10. I would like to know the proper setup of all blocks in none pretrained weights settings.

    opened by Aminullah6264 0
Owner
蒋子航
Now a Ph.D. student supervised by Prof. Feng Jiashi in ECE, NUS.
蒋子航
Implementation of QuickDraw - an online game developed by Google, combined with AirGesture - a simple gesture recognition application

QuickDraw - AirGesture Introduction Here is my python source code for QuickDraw - an online game developed by google, combined with AirGesture - a sim

Viet Nguyen 89 Dec 18, 2022
School of Artificial Intelligence at the Nanjing University (NJU)School of Artificial Intelligence at the Nanjing University (NJU)

F-Principle This is an exercise problem of the digital signal processing (DSP) course at School of Artificial Intelligence at the Nanjing University (

Thyrix 5 Nov 23, 2022
Resources related to our paper "CLIN-X: pre-trained language models and a study on cross-task transfer for concept extraction in the clinical domain"

CLIN-X (CLIN-X-ES) & (CLIN-X-EN) This repository holds the companion code for the system reported in the paper: "CLIN-X: pre-trained language models a

Bosch Research 4 Dec 05, 2022
[ICCV '21] In this repository you find the code to our paper Keypoint Communities

Keypoint Communities In this repository you will find the code to our ICCV '21 paper: Keypoint Communities Duncan Zauss, Sven Kreiss, Alexandre Alahi,

Duncan Zauss 262 Dec 13, 2022
PyTorch Code of "Memory In Memory: A Predictive Neural Network for Learning Higher-Order Non-Stationarity from Spatiotemporal Dynamics"

Memory In Memory Networks It is based on the paper Memory In Memory: A Predictive Neural Network for Learning Higher-Order Non-Stationarity from Spati

Yang Li 12 May 30, 2022
PyTorch implementation of the Transformer in Post-LN (Post-LayerNorm) and Pre-LN (Pre-LayerNorm).

Transformer-PyTorch A PyTorch implementation of the Transformer from the paper Attention is All You Need in both Post-LN (Post-LayerNorm) and Pre-LN (

Jared Wang 22 Feb 27, 2022
Multi-task Multi-agent Soft Actor Critic for SMAC

Multi-task Multi-agent Soft Actor Critic for SMAC Overview The CARE formulti-task: Multi-Task Reinforcement Learning with Context-based Representation

RuanJingqing 8 Sep 30, 2022
Hunt down social media accounts by username across social networks

Hunt down social media accounts by username across social networks Installation | Usage | Docker Notes | Contributing Installation # clone the repo $

1 Dec 14, 2021
Facebook AI Image Similarity Challenge: Descriptor Track

Facebook AI Image Similarity Challenge: Descriptor Track This repository contains the code for our solution to the Facebook AI Image Similarity Challe

Sergio MP 17 Dec 14, 2022
Decorators for maximizing memory utilization with PyTorch & CUDA

torch-max-mem This package provides decorators for memory utilization maximization with PyTorch and CUDA by starting with a maximum parameter size and

Max Berrendorf 10 May 02, 2022
PyTorch code of "SLAPS: Self-Supervision Improves Structure Learning for Graph Neural Networks"

SLAPS-GNN This repo contains the implementation of the model proposed in SLAPS: Self-Supervision Improves Structure Learning for Graph Neural Networks

60 Dec 22, 2022
SegTransVAE: Hybrid CNN - Transformer with Regularization for medical image segmentation

SegTransVAE: Hybrid CNN - Transformer with Regularization for medical image segmentation This repo is the official implementation for SegTransVAE. Seg

Nguyen Truong Hai 4 Aug 04, 2022
Keqing Chatbot With Python

KeqingChatbot A public running instance can be found on telegram as @keqingchat_bot. Requirements Python 3.8 or higher. A bot token. Local Deploy git

Rikka-Chan 2 Jan 16, 2022
Neural Network to colorize grayscale images

#colornet Neural Network to colorize grayscale images Results Grayscale Prediction Ground Truth Eiji K used colornet for anime colorization Sources Au

Pavel Hanchar 3.6k Dec 24, 2022
AITom is an open-source platform for AI driven cellular electron cryo-tomography analysis.

AITom Introduction AITom is an open-source platform for AI driven cellular electron cryo-tomography analysis. AITom is originated from the tomominer l

93 Jan 02, 2023
Invariant Causal Prediction for Block MDPs

MISA Abstract Generalization across environments is critical to the successful application of reinforcement learning algorithms to real-world challeng

Meta Research 41 Sep 17, 2022
R interface to fast.ai

R interface to fastai The fastai package provides R wrappers to fastai. The fastai library simplifies training fast and accurate neural nets using mod

113 Dec 20, 2022
Pseudo-rng-app - whos needs science to make a random number when you have pseudoscience?

Pseudo-random numbers with pseudoscience rng is so complicated! Why cant we have a horoscopic, vibe-y way of calculating a random number? Why cant rng

Andrew Blance 1 Dec 27, 2021
Code for TIP 2017 paper --- Illumination Decomposition for Photograph with Multiple Light Sources.

Illumination_Decomposition Code for TIP 2017 paper --- Illumination Decomposition for Photograph with Multiple Light Sources. This code implements the

QAY 7 Nov 15, 2020
Code accompanying the paper Say As You Wish: Fine-grained Control of Image Caption Generation with Abstract Scene Graphs (Chen et al., CVPR 2020, Oral).

Say As You Wish: Fine-grained Control of Image Caption Generation with Abstract Scene Graphs This repository contains PyTorch implementation of our pa

Shizhe Chen 178 Dec 29, 2022