https://github.com/acecoooool/segmentation
Semantic Segmentation sub-project
https://github.com/acecoooool/segmentation
pytorch segmentation semantic-segmentation state-of-the-art
Last synced: 3 months ago
JSON representation
Semantic Segmentation sub-project
- Host: GitHub
- URL: https://github.com/acecoooool/segmentation
- Owner: AceCoooool
- License: mit
- Created: 2019-04-27T00:42:49.000Z (over 6 years ago)
- Default Branch: master
- Last Pushed: 2019-05-15T11:21:25.000Z (over 6 years ago)
- Last Synced: 2025-06-01T12:59:24.892Z (4 months ago)
- Topics: pytorch, segmentation, semantic-segmentation, state-of-the-art
- Language: Python
- Homepage:
- Size: 174 KB
- Stars: 2
- Watchers: 2
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Semantic Segmentation
This is a sub-project of [pytorch-cv](https://github.com/AceCoooool/pytorch-cv)（for convenient）
**Support Models：**
- [x] FCN
- [x] PSPNet
- [x] DeepLabv3
- [x] DANet
- [x] OCNet
## Environment
- PyTorch 1.1
## Performance
#### Pascal VOC 2012
Here, we using train (10582), val (1449), test (1456) as most paper used. (More detail can reference [DeepLabv3](https://github.com/chenxi116/DeepLabv3.pytorch)) . And the performance is evaluated with single scale
- Base Size 540, Crop Size 480
| Model | backbone | Paper | OHEM | aux | dilated | JPU | Epoch | val (crop) | val |
| :--------: | :-----------: | :--------: | :--: | :--: | :-----: | :--: | :---: | :----------------------------------------------------------: | :---------: |
| FCN | ResNet101-v1s | / | ✗ | ✓ | ✗ | ✓ | 50 | [94.54/78.31](https://drive.google.com/open?id=1-FF5BUSB9hNCyldC1nV35LeWLSQFa9Jl) | 94.50/76.89 |
| PSPNet | ResNet101-v1s | / | ✗ | ✓ | ✓ | ✗ | 50 | [94.87/80.13](https://drive.google.com/open?id=1g40cVTJRCHLBKwVqjev4yQA5YflQwhrv) | 94.88/78.57 |
| PSPNet | ResNet101-v1s | / | ✗ | ✓ | ✗ | ✓ | 50 | [94.89/79.90](https://drive.google.com/open?id=1XRFiijt0tAbLgXhV5oVEot6qLccz9JyA) | 94.77/78.48 |
| DeepLabv3 | ResNet101-v1s | no / 77.02 | ✗ | ✓ | ✗ | ✓ | 50 | [95.17/81.00](https://drive.google.com/open?id=1R0C6qwCxOLztps4odVWLiZpe57n5TuDX) | 94.81/78.75 |
| DANet | ResNet101-v1s | / | ✗ | ✓ | ✗ | ✓ | 50 | [94.98/80.49](https://drive.google.com/open?id=1jS69enf_dyn8l27DfMDuzt_ulva7KSWt) | 94.85/78.72 |
| OCNet-Base | ResNet101-v1s | / | ✗ | ✓ | ✗ | ✓ | 50 | [94.91/80.33](https://drive.google.com/open?id=15gs_gzgAT_hciPgwm12G_MRMi0VZg9Gb) | 94.86/79.07 |
| OCNet-ASP | ResNet101-v1s | / | ✗ | ✓ | ✗ | ✓ | 50 | | |
> 1. the metric is `pixAcc/mIoU`
> 2. `aux_weight=0.5`
#### Cityscapes
Here, we only using fine train (2975), val (500) as most paper used. (More detail can reference [DeepLabv3](https://github.com/chenxi116/DeepLabv3.pytorch)) . And the performance is evaluated with single scale
- Base Size 1024, Crop Size 768
| Model | backbone | Paper(*) | OHEM | aux | dilated | JPU | Epoch | val (crop) | val |
| :--------: | :-----------: | :------: | :--: | :--: | :-----: | :--: | :---: | :---------: | :----------------------------------------------------------: |
| FCN | ResNet101-v1s | no/75.96 | ✗ | ✓ | ✗ | ✓ | 120 | 96.29/73.60 | [96.18/78.61](https://drive.google.com/open?id=119bPxJwL6zAtEyvEJiHljK6US9f-d5Xr) |
| PSPNet | ResNet101-v1s | no/78.56 | ✗ | ✓ | ✗ | ✓ | 120 | 96.21/73.64 | [96.09/78.62](https://drive.google.com/open?id=1qJVXevkErgVvsa4x8mCvAdMlCIuVMJ1C) |
| DeepLabv3 | ResNet101-v1s | no/78.90 | ✗ | ✓ | ✗ | ✓ | 120 | 96.25/73.44 | [96.23/79.03](https://drive.google.com/open?id=1_XIIeIKEMbg4M2SO49Vq756d20zjeXxk) |
| DANet | ResNet101-v1s | no/78.83 | | | | | | | |
| OCNet-Base | ResNet101-v1s | no/79.67 | ✗ | ✓ | ✗ | ✓ | 120 | 96.30/74.18 | [TODO](https://drive.google.com/open?id=1q2h8d_mJeyHfmCRQellTSpralv8ubI5L) |
| OCNet-ASP | ResNet101-v1s | | | | | | | | |
> Note：
>
> 1. Paper(*) means results from: [openseg.pytorch](https://github.com/openseg-group/openseg.pytorch)（results with single scale without crop），there are a little different in the training strategy.
## Demo
Demo of segmentation of a given image. (Please download pre-trained model to `~/.torch/models` first. --- If you put pre-trained model to other folder, please change the `--root`)
```shell
$ python demo_segmentation_pil.py [--model fcn_resnet101_voc] [--input-pic .jpg] [--cuda true] [--aux true] [--jpu true] [--dilated false]
```
> Note：
>
> 1. if not give `--input-pic`, using default image we provided.
> 2. `aux, jpu, dilated` is depend on your model
## Evaluation
The default data root is `~/.torch/datasets` (You can download dataset and build a soft-link to it)
```shell
$ python eval_segmentation_pil.py [--model_name fcn_resnet101_voc] [--dataset pascal_paper] [--split val] [--mode testval|val] [--base-size 540] [--crop-size 480] [--aux true] [--jpu true] [--dilated false] [--cuda true]
```
> Note：
>
> 1. if you choose `mode=testval`，you can not set `base-size` and `crop-size`
> 2. `aux, jpu, dilated` is depend on your model
## Train
Download pre-trained backbone and put it on `~/.torch/models`
Recommend to using distributed training.
```shell
$ export NGPUS=4
$ python -m torch.distributed.launch --nproc_per_node=$NGPUS train_segmentation_pil.py [--model fcn] [--backbone resnet101] [--dataset pascal_voc] [--batch-size 8] [--base-size 540] [--crop-size 480] [--aux true] [--jpu true] [--dilated false] [--log-step 10]
```
> Our training results' setting can see [train.sh](scripts/train.sh)
## Prepare data
#### VOC2012
```shell
mkdir data
cd data
wget http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar
tar -xf VOCtrainval_11-May-2012.tar
cd VOCdevkit/VOC2012/
wget http://cs.jhu.edu/~cxliu/data/SegmentationClassAug.zip
wget http://cs.jhu.edu/~cxliu/data/SegmentationClassAug_Visualization.zip
wget http://cs.jhu.edu/~cxliu/data/list.zip
unzip SegmentationClassAug.zip
unzip SegmentationClassAug_Visualization.zip
unzip list.zip
```
Your can make a soft link to `.torch/datasets/voc`
#### Cityscapes
```shell
unzip leftImg8bit_trainvaltest.zip
unzip gtFine_trainvaltest.zip
git clone https://github.com/mcordts/cityscapesScripts.git
mv cityscapesScripts/cityscapesscripts ./
```
Your can make a soft link to `.torch/datasets/citys`
## Download
#### Backbone
| resnet50-v1s | resnet101-v1s |
| ------------------------------------------------------------ | ------------------------------------------------------------ |
| [GoogleDrive](https://drive.google.com/open?id=1Mx_SIv1o1qjRz1tqEc-ggQ_MtZKQT3ET) | [GoogleDrive](https://drive.google.com/open?id=1pA_tN2MFi-7J5n1og10kDnV8raphM3V-) |