https://github.com/jfzhuang/IFR

[CVPR'22] Semi-Supervised Video Semantic Segmentation with Inter-Frame Feature Reconstruction
https://github.com/jfzhuang/IFR

cvpr2022 semantic-segmentation semi-supervised-learning semi-supervised-segmentation

Last synced: 2 months ago
JSON representation

[CVPR'22] Semi-Supervised Video Semantic Segmentation with Inter-Frame Feature Reconstruction

• Host: GitHub
• URL: https://github.com/jfzhuang/IFR
• Owner: jfzhuang
• License: mit
• Created: 2021-11-22T07:53:14.000Z (almost 3 years ago)
• Default Branch: main
• Last Pushed: 2022-10-17T10:29:53.000Z (almost 2 years ago)
• Last Synced: 2024-06-15T00:48:53.860Z (4 months ago)
• Topics: cvpr2022, semantic-segmentation, semi-supervised-learning, semi-supervised-segmentation
• Language: Python
• Homepage:
• Size: 19.7 MB
• Stars: 27
• Watchers: 2
• Forks: 3
• Open Issues: 3
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• Awesome-Semi-Supervised-Semantic-Segmentation - Code - Supervised_Video_Semantic_Segmentation_With_Inter-Frame_Feature_Reconstruction_CVPR_2022_paper.pdf)| (2022)

README

        
# IFR

This repository is the official implementation of "Semi-Supervised Video Semantic Segmentation with Inter-Frame Feature Reconstruction" (accepted by CVPR 2022). It is designed for semi-supervised video semantic segmentation task.

## Install & Requirements

The code has been tested on pytorch=1.8.2 and python3.8. Please refer to `requirements.txt` for detailed information.

**To Install python packages**

```

pip install -r requirements.txt

```

## Download Pretrained Weights

````bash

mkdir ./IFR/pretrained

cd ./IFR/pretrained

# download resnet18 imagenet pretrained weight

wget http://sceneparsing.csail.mit.edu/model/pretrained_resnet/resnet18-imagenet.pth

# download resnet101 imagenet pretrained weight

wget http://sceneparsing.csail.mit.edu/model/pretrained_resnet/resnet101-imagenet.pth

````

## Data preparation

You need to download the [Cityscapes](https://www.cityscapes-dataset.com/) datasets.

Your directory tree should be look like this:

````bash

./IFR/data

├── cityscapes

│   ├── gtFine

│   │   ├── train

│   │   └── val

│   └── leftImg8bit_sequence

│       ├── train

│       └── val

````

## Prepare Downsample Dataset

Generated downsample dataset would be saved in ./data

````bash

cd ./IFR

python tools/data_downsample.py

````

## Stage One Training of Accel

For example, train image segmentation model on 2 GPUs. Checkpoints would be saved in ./IFR/work_dirs.

````bash

# train PSP18 baseline model

cd ./IFR/exp/sup_30_res18/scripts

bash train.sh

# train PSP101 baseline model

cd ./IFR/exp/sup_30_res101/scripts

bash train.sh

# train PSP18 IFR model

cd ./IFR/exp/IFR_30_res18/scripts

bash train.sh

# train PSP101 IFR model

cd ./IFR/exp/IFR_30_res101/scripts

bash train.sh

````

## Stage Two Training of Accel

For example, train Accel18 on 2 GPUs. Checkpoints would be saved in ./Accel/work_dirs.

````bash

mkdir ./Accel/work_dirs

# train Accel18 with baseline model

cd ./Accel/exp/accel18_30_sup/script

bash train.sh

# train Accel18 with IFR model

cd ./Accel/exp/accel18_30_IFR/script

bash train.sh

````