Skip to content

Official PyTorch implementation of GroupViT: Semantic Segmentation Emerges from Text Supervision, CVPR 2022.

License

Notifications You must be signed in to change notification settings

NVlabs/GroupViT

Folders and files

NameName
Last commit message
Last commit date

Latest commit

Β 

History

19 Commits
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 

Repository files navigation

GroupViT: Semantic Segmentation Emerges from Text Supervision

GroupViT is a framework for learning semantic segmentation purely from text captions without using any mask supervision. It learns to perform bottom-up heirarchical spatial grouping of semantically-related visual regions. This repository is the official implementation of GroupViT introduced in the paper:

GroupViT: Semantic Segmentation Emerges from Text Supervision, Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, Xiaolong Wang, CVPR 2022.

Visual Results

Links

Citation

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

@article{xu2022groupvit,
  author    = {Xu, Jiarui and De Mello, Shalini and Liu, Sifei and Byeon, Wonmin and Breuel, Thomas and Kautz, Jan and Wang, Xiaolong},
  title     = {GroupViT: Semantic Segmentation Emerges from Text Supervision},
  journal   = {arXiv preprint arXiv:2202.11094},
  year      = {2022},
}

Environmental Setup

  • Python 3.7
  • PyTorch 1.8
  • webdataset 0.1.103
  • mmsegmentation 0.18.0
  • timm 0.4.12

Instructions:

conda create -n groupvit python=3.7 -y
conda activate groupvit
conda install pytorch==1.8.0 torchvision==0.9.0 cudatoolkit=11.1 -c pytorch -c conda-forge
pip install mmcv-full==1.3.14 -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.8.0/index.html
pip install mmsegmentation==0.18.0
pip install webdataset==0.1.103
pip install timm==0.4.12
git clone https://github.com/NVIDIA/apex
cd && apex && pip install -v --disable-pip-version-check --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./
pip install opencv-python==4.4.0.46 termcolor==1.1.0 diffdist einops omegaconf
pip install nltk ftfy regex tqdm

Demo

python demo/demo_seg.py --cfg configs/group_vit_gcc_yfcc_30e.yml --resume /path/to/checkpoint --vis input_pred_label final_group --input demo/examples/voc.jpg --output_dir demo/output

The output is saved in demo/output/.

Benchmark Results

Zero-shot Classification Zero-shot Segmentation
config ImageNet Pascal VOC Pascal Context COCO
GCC + YFCC (cfg) 43.7 52.3 22.4 24.3
GCC + RedCaps (cfg) 51.6 50.8 23.7 27.5

Pre-trained weights group_vit_gcc_yfcc_30e-879422e0.pth and group_vit_gcc_redcap_30e-3dd09a76.pth for these models are provided by Jiarui Xu here.

Data Preparation

During training, we use webdataset for scalable data loading. To convert image text pairs into the webdataset format, we use the img2dataset tool to download and preprocess the dataset.

For inference, we use mmsegmentation for semantic segmentation testing, evaluation and visualization on Pascal VOC, Pascal Context and COCO datasets.

The overall file structure is as follows:

GroupViT
β”œβ”€β”€ local_data
β”‚   β”œβ”€β”€ gcc3m_shards
β”‚   β”‚   β”œβ”€β”€ gcc-train-000000.tar
β”‚   β”‚   β”œβ”€β”€ ...
β”‚   β”‚   β”œβ”€β”€ gcc-train-000436.tar
β”‚   β”œβ”€β”€ gcc12m_shards
β”‚   β”‚   β”œβ”€β”€ gcc-conceptual-12m-000000.tar
β”‚   β”‚   β”œβ”€β”€ ...
β”‚   β”‚   β”œβ”€β”€ gcc-conceptual-12m-001943.tar
β”‚   β”œβ”€β”€ yfcc14m_shards
β”‚   β”‚   β”œβ”€β”€ yfcc14m-000000.tar
β”‚   β”‚   β”œβ”€β”€ ...
β”‚   β”‚   β”œβ”€β”€ yfcc14m-001888.tar
β”‚   β”œβ”€β”€ redcap12m_shards
β”‚   β”‚   β”œβ”€β”€ redcap12m-000000.tar
β”‚   β”‚   β”œβ”€β”€ ...
β”‚   β”‚   β”œβ”€β”€ redcap12m-001211.tar
β”‚   β”œβ”€β”€ imagenet_shards
β”‚   β”‚   β”œβ”€β”€ imagenet-val-000000.tar
β”‚   β”‚   β”œβ”€β”€ ...
β”‚   β”‚   β”œβ”€β”€ imagenet-val-000049.tar
β”‚   β”œβ”€β”€ VOCdevkit
β”‚   β”‚   β”œβ”€β”€ VOC2012
β”‚   β”‚   β”‚   β”œβ”€β”€ JPEGImages
β”‚   β”‚   β”‚   β”œβ”€β”€ SegmentationClass
β”‚   β”‚   β”‚   β”œβ”€β”€ ImageSets
β”‚   β”‚   β”‚   β”‚   β”œβ”€β”€ Segmentation
β”‚   β”‚   β”œβ”€β”€ VOC2010
β”‚   β”‚   β”‚   β”œβ”€β”€ JPEGImages
β”‚   β”‚   β”‚   β”œβ”€β”€ SegmentationClassContext
β”‚   β”‚   β”‚   β”œβ”€β”€ ImageSets
β”‚   β”‚   β”‚   β”‚   β”œβ”€β”€ SegmentationContext
β”‚   β”‚   β”‚   β”‚   β”‚   β”œβ”€β”€ train.txt
β”‚   β”‚   β”‚   β”‚   β”‚   β”œβ”€β”€ val.txt
β”‚   β”‚   β”‚   β”œβ”€β”€ trainval_merged.json
β”‚   β”‚   β”œβ”€β”€ VOCaug
β”‚   β”‚   β”‚   β”œβ”€β”€ dataset
β”‚   β”‚   β”‚   β”‚   β”œβ”€β”€ cls
β”‚   β”œβ”€β”€ coco
β”‚   β”‚   β”œβ”€β”€ images
β”‚   β”‚   β”‚   β”œβ”€β”€ train2017
β”‚   β”‚   β”‚   β”œβ”€β”€ val2017
β”‚   β”‚   β”œβ”€β”€ annotations
β”‚   β”‚   β”‚   β”œβ”€β”€ train2017
β”‚   β”‚   β”‚   β”œβ”€β”€ val2017

The instructions for preparing each dataset are as follows.

GCC3M

Please download the training split annotation file from Conceptual Caption 12M and name it as gcc3m.tsv.

Then run img2dataset to download the image text pairs and save them in the webdataset format.

sed -i '1s/^/caption\turl\n/' gcc3m.tsv
img2dataset --url_list gcc3m.tsv --input_format "tsv" \
            --url_col "url" --caption_col "caption" --output_format webdataset\
            --output_folder local_data/gcc3m_shards
            --processes_count 16 --thread_count 64
            --image_size 512 --resize_mode keep_ratio --resize_only_if_bigger True \
            --enable_wandb True --save_metadata False --oom_shard_count 6
rename -d 's/^/gcc-train-/' local_data/gcc3m_shards/*

Please refer to img2dataset CC3M tutorial for more details.

GCC12M

Please download the annotation file from Conceptual Caption 12M and name it as gcc12m.tsv.

Then run img2dataset to download the image text pairs and save them in the webdataset format.

sed -i '1s/^/caption\turl\n/' gcc12m.tsv
img2dataset --url_list gcc12m.tsv --input_format "tsv" \
            --url_col "url" --caption_col "caption" --output_format webdataset\
            --output_folder local_data/gcc12m_shards \
            --processes_count 16 --thread_count 64
            --image_size 512 --resize_mode keep_ratio --resize_only_if_bigger True \
            --enable_wandb True --save_metadata False --oom_shard_count 6
rename -d 's/^/gcc-conceptual-12m-/' local_data/gcc12m_shards/*

Please refer to img2dataset CC12M tutorial for more details.

YFCC14M

Please follow the CLIP Data Preparation instructions to download the YFCC14M subset.

wget https://openaipublic.azureedge.net/clip/data/yfcc100m_subset_data.tsv.bz2
bunzip2 yfcc100m_subset_data.tsv.bz2

Then run the preprocessing script to create the subset sql db and annotation tsv files. This may take a while.

python convert_dataset/create_subset.py --input-dir . --output-dir . --subset yfcc100m_subset_data.tsv

This script will create two files: an SQLite db called yfcc100m_dataset.sql and an annotation tsv file called yfcc14m_dataset.tsv.

Then follow the YFCC100M Download Instruction to download the dataset and its metadata file.

pip install git+https://gitlab.com/jfolz/yfcc100m.git
mkdir -p yfcc100m_meta
python -m yfcc100m.convert_metadata . -o yfcc100m_meta --skip_verification
mkdir -p yfcc100m_zip
python -m yfcc100m.download yfcc100m_meta -o yfcc100m_zip

Finally convert the dataset into the webdataset format.

python convert_dataset/convert_yfcc14m.py --root yfcc100m_zip --info yfcc14m_dataset.tsv --shards yfcc14m_shards

RedCaps12M

Please download the annotation file from RedCaps.

wget https://www.dropbox.com/s/cqtdpsl4hewlli1/redcaps_v1.0_annotations.zip?dl=1
unzip redcaps_v1.0_annotations.zip

Then run the preprocessing script and img2dataset to download the image text pairs and save them in the webdataset format.

python convert_dataset/process_redcaps.py annotations redcaps12m_meta/redcaps12m.parquet --num-split 16
img2dataset --url_list ~/data/redcaps12m/ --input_format "parquet" \
            --url_col "URL" --caption_col "TEXT" --output_format webdataset \
            --output_folder local_data/recaps12m_shards
            --processes_count 16 --thread_count 64
            --image_size 512 --resize_mode keep_ratio --resize_only_if_bigger True \
            --enable_wandb True --save_metadata False --oom_shard_count 6
rename -d 's/^/redcap12m-/' local_data/recaps12m_shards/*

ImageNet

Please follow the webdataset ImageNet Example to convert ImageNet into the webdataset format.

Pascal VOC

Please follow the MMSegmentation Pascal VOC Preparation instructions to download and setup the Pascal VOC dataset.

Pascal Context

Please refer to the MMSegmentation Pascal Context Preparation instructions to download and setup the Pascal Context dataset.

COCO

COCO dataset is an object detection dataset with instance segmentation annotations. To evaluate GroupViT, we combine all the instance masks of a catergory together and generate semantic segmentation maps. To generate the semantic segmentation maps, please follow MMSegmentation's documentation to download the COCO-Stuff-164k dataset first and then run the following

python convert_dataset/convert_coco.py local_data/data/coco/ -o local_data/data/coco/

Run Experiments

Pre-train

Train on a single node:

(node0)$ ./tools/dist_launch.sh main_group_vit.py /path/to/config $GPUS_PER_NODE

For example, to train on a node with 8 GPUs, run:

(node0)$ ./tools/dist_launch.sh main_group_vit configs/group_vit_gcc_yfcc_30e.yml 8

Train on multiple nodes:

(node0)$ ./tools/dist_mn_launch.sh main_group_vit.py /path/to/config $NODE_RANK $NUM_NODES $GPUS_PER_NODE $MASTER_ADDR
(node1)$ ./tools/dist_mn_launch.sh main_group_vit.py /path/to/config $NODE_RANK $NUM_NODES $GPUS_PER_NODE $MASTER_ADDR

For example, to train on two nodes with 8 GPUs each, run:

(node0)$ ./tools/dist_mn_launch.sh main_group_vit.py configs/group_vit_gcc_yfcc_30e.yml 0 2 8 tcp://node0
(node1)$ ./tools/dist_mn_launch.sh main_group_vit.py configs/group_vit_gcc_yfcc_30e.yml 1 2 8 tcp://node0

We used 16 NVIDIA V100 GPUs for pre-training (in 2 days) in our paper.

Zero-shot Transfer to Image Classification

ImageNet

./tools/dist_launch.sh main_group_vit.py /path/to/config $NUM_GPUS --resume /path/to/checkpoint --eval

Zero-shot Transfer to Semantic Segmentation

Pascal VOC

./tools/dist_launch.sh main_seg.py /path/to/config $NUM_GPUS --resume /path/to/checkpoint

Pascal Context

./tools/dist_launch.sh main_seg.py /path/to/config $NUM_GPUS --resume /path/to/checkpoint --opts evaluate.seg.cfg segmentation/configs/_base_/datasets/pascal_context.py

COCO

./tools/dist_launch.sh main_seg.py /path/to/config $NUM_GPUS --resume /path/to/checkpoint --opts evaluate.seg.cfg segmentation/configs/_base_/datasets/coco.py