KO | EN

GitLite

Minimalist Open Source Repository Search Engine

gitlite — search

#developer-tools #typescript #claude-code #codex #dsh #dsh-plugin #ai-agent #ai-agents #deepseek-harness #agent-skills #macos #cli

DKD

★ 68 Open GitHub ↗

An official implementation of "Decomposed Knowledge Distillation for Class-incremental Semantic Segmentation" (NeurIPS 2022) in PyTorch.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Decomposed Knowledge Distillation for Class-Incremental Semantic Segmentation

This is an official implementation of the paper "Decomposed Knowledge Distillation for Class-Incremental Semantic Segmentation", accepted to NeurIPS 2022.

For more information, please checkout the project site [[website](https://cvlab.yonsei.ac.kr/projects/DKD/)] and our paper [[arXiv](http://arxiv.org/abs/2210.05941) / [OpenReview](https://openreview.net/forum?id=0SgKq4ZC9r)].

## Pre-requisites
This repository has been tested with the following libraries:
* Python (3.6)
* Pytorch (1.8.1)

## Getting Started

### Datasets
#### PASCAL VOC 2012
We use augmented 10,582 training samples and 1,449 validation samples for PASCAL VOC 2012. You can download the original dataset in [here](http://host.robots.ox.ac.uk/pascal/VOC/voc2012/index.html#devkit). To train our model with augmented samples, please download labels of augmented samples (['SegmentationClassAug'](https://www.dropbox.com/s/oeu149j8qtbs1x0/SegmentationClassAug.zip)) and file names (['train_aug.txt'](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/train_aug.txt)). The structure of data path should be organized as follows:
```bash
└── /dataset/VOC2012
    ├── Annotations
    ├── ImageSets
    │   └── Segmentation
    │       ├── train_aug.txt
    │       └── val.txt
    ├── JPEGImages
    ├── SegmentationClass
    └── SegmentationClassAug
```

#### ADE20K
We use 20,210 training samples and 2,000 validation samples for ADE20K. You can download the dataset in [here](http://sceneparsing.csail.mit.edu/). The structure of data path should be organized as follows:
```bash
└── /dataset/ADEChallengeData2016
    ├── annotations
    ├── images
    ├── objectInfo150.txt
    └── sceneCategories.txt
```

### Training
#### PASCAL VOC 2012
```Shell
# An example srcipt for 15-5 overlapped setting of PASCAL VOC

GPU=0,1,2,3
BS=8  # Total 32
SAVEDIR='saved_voc'

TASKSETTING='overlap'  # or 'disjoint'
TASKNAME='15-5' # or ['15-1', '19-1', '10-1', '5-3']
EPOCH=60
INIT_LR=0.001
LR=0.0001
INIT_POSWEIGHT=2
MEMORY_SIZE=0  # 100 for DKD-M

NAME='DKD'
python train_voc.py -c configs/config_voc.json \
-d ${GPU} --multiprocessing_distributed --save_dir ${SAVEDIR} --name ${NAME} \
--task_name ${TASKNAME} --task_setting ${TASKSETTING} --task_step 0 --lr ${INIT_LR} --bs ${BS} --pos_weight ${INIT_POSWEIGHT}

python train_voc.py -c configs/config_voc.json \
-d ${GPU} --multiprocessing_distributed --save_dir ${SAVEDIR} --name ${NAME} \
--task_name ${TASKNAME} --task_setting ${TASKSETTING} --task_step 1 --lr ${LR} --bs ${BS} --freeze_bn --mem_size ${MEMORY_SIZE}
```

#### ADE20K
```Shell
# An example srcipt for 50-50 overlapped setting of ADE20K

GPU=0,1,2,3
BS=6  # Total 24
SAVEDIR='saved_ade'

TASKSETTING='overlap'
TASKNAME='50-50' # or ['100-10', '100-50']
EPOCH=100
INIT_LR=0.0025
LR=0.00025
MEMORY_SIZE=0 # 300 for DKD-M

NAME='DKD'
python train_ade.py -c configs/config_ade.json \
-d ${GPU} --multiprocessing_distributed --save_dir ${SAVEDIR} --name ${NAME} \
--task_name ${TASKNAME} --task_setting ${TASKSETTING} --task_step 0 --lr ${INIT_LR} --bs ${BS}

python train_ade.py -c configs/config_ade.json \
-d ${GPU} --multiprocessing_distributed --save_dir ${SAVEDIR} --name ${NAME} \
--task_name ${TASKNAME} --task_setting ${TASKSETTING} --task_step 1 --lr ${LR} --bs ${BS} --freeze_bn --mem_size ${MEMORY_SIZE}

python train_ade.py -c configs/config_ade.json \
-d ${GPU} --multiprocessing_distributed --save_dir ${SAVEDIR} --name ${NAME} \
--task_name ${TASKNAME} --task_setting ${TASKSETTING} --task_step 2 --lr ${LR} --bs ${BS} --freeze_bn --mem_size ${MEMORY_SIZE}
```

### Testing
#### PASCAL VOC 2012
```Shell
python eval_voc.py -d 0 -r path/to/weight.pth
```
We provide pretrained weights and configuration files. The results should be:
| Method (Overlapped) | VOC 19-1 (2 steps) $\text{hIoU}$ / $\text{mIoU}_{\text{all}}$ | VOC 15-5 (2 steps) $\text{hIoU}$ / $\text{mIoU}_{\text{all}}$ | VOC 15-1 (6 steps) $\text{hIoU}$ / $\text{mIoU}_{\text{all}}$ | VOC 10-1 (11 steps) $\text{hIoU}$ / $\text{mIoU}_{\text{all}}$ | VOC 5-3 (6 steps) $\text{hIoU}$ / $\text{mIoU}_{\text{all}}$ |
| :- | :-----------: | :-----------: | :-----------: | :-----------: | :-----------: |
| DKD | [56.95 / 76.13](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/voc_overlapped_19-1.zip) | [67.17 / 73.95](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/voc_overlapped_15-1.zip) | [57.46 / 70.50](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/voc_overlapped_15-1.zip) | [57.21 / 60.43](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/voc_overlapped_10-1.zip) | [61.32 / 58.98](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/voc_overlapped_5-3.zip) |
| DKD-M | [68.14 / 77.04](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/voc_overlapped_19-1_memory.zip) | [68.96 / 74.84](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/voc_overlapped_15-5_memory.zip) | [64.09 / 72.95](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/voc_overlapped_15-1_memory.zip) | [64.89 / 66.20](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/voc_overlapped_10-1_memory.zip) | [65.02 / 63.32](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/voc_overlapped_5_3_memory.zip) |

| Method (Disjoint) | VOC 19-1 (2 steps) $\text{hIoU}$ / $\text{mIoU}_{\text{all}}$ | VOC 15-5 (2 steps) $\text{hIoU}$ / $\text{mIoU}_{\text{all}}$ | VOC 15-1 (6 steps) $\text{hIoU}$ / $\text{mIoU}_{\text{all}}$ | 
| :- | :-----------: | :-----------: | :-----------: |
| DKD | [57.98 / 75.90](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/voc_disjoint_19-1.zip) | [64.35 / 72.21](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/voc_disjoint_15-1.zip) | [54.21 / 68.48](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/voc_disjoint_15-1.zip) |
| DKD-M | [67.12 / 76.84](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/voc_disjoint_19-1_memory.zip) | [65.04 / 72.53](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/voc_disjoint_15-5_memory.zip) | [60.32 / 70.78](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/voc_disjoint_15-1_memory.zip) |

#### ADE20K
```Shell
python eval_ade.py -d 0 -r path/to/weight.pth
```
We provide pretrained weights and configuration files. The results should be:
| Method | ADE 100-50 (2 steps) $\text{hIoU}$ / $\text{mIoU}_{\text{all}}$ | ADE 100-10 (6 steps) $\text{hIoU}$ / $\text{mIoU}_{\text{all}}$ | ADE 50-50 (3 steps) $\text{hIoU}$ / $\text{mIoU}_{\text{all}}$ |
| :---- | :-----------: | :-----------: | :-----------: |
| DKD | [29.42 / 35.55](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/ade_overlapped_100-50.zip) | [26.79 / 34.53](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/ade_overlapped_100-10.zip) | [34.77 / 34.39](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/ade_overlapped_50-50.zip) |
| DKD-M | [29.45 / 35.58](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/ade_overlapped_100-50_memory.zip) | [27.44 / 34.83](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/ade_overlapped_100-10.zip) | [34.78 / 34.39](https://github.com/cvlab-yonsei/DKD/releases/download/v1.0/ade_overlapped_50-50.zip) |

## Acknowledgements
* This template is borrowed from [pytorch-template](https://github.com/victoresque/pytorch-template).

DKD

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README