This repo contains the code for my experiments on mask completion using the PCNet-M model proposed in Self-Supervised Scene De-occlusion.
- Clone the repo:
git clone https://github.com/praeclarumjj3/PCNetM-Experiments.git
cd PCNetM-Experiments
- Install pycocotools:
pip install "git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI"
- Install Pytorch and other dependencies:
pip3 install -r requirements.txt
- Download the MS-COCO 2014 images and unzip:
wget http://images.cocodataset.org/zips/train2014.zip
wget http://images.cocodataset.org/zips/val2014.zip
- Download the annotations and untar:
gdown https://drive.google.com/uc?id=0B8e3LNo7STslZURoTzhhMFpCelE
tar -xf annotations.tar.gz
- Unzip the files according to the following structure
PCNetM-Experiments
├── data
│ ├── COCOA
│ │ ├── annotations
│ │ ├── train2014
│ │ ├── val2014
-
Download released models here and put the folder
released
underPCNetM-Experiments
. -
Run
demos/demo_cocoa.ipynb
. There are some test examples fordemos/demo_cocoa.ipynb
in the repo, so you don't have to download the COCOA dataset if you just want to try a few samples. -
If you want to use predicted modal masks by existing instance segmentation models, you need to adjust some parameters in the demo, please refer to the answers in this issue.
- Run the following command:
sh experiments/COCOA/pcnet_m/train.sh # you may have to set--nproc_per_node=#YOUR_GPUS
Best Loss: 0.0674 after 44000 iterations.
- Execute:
sh tools/test_cocoa.sh
Metric | Value |
---|---|
acc_allpair | 0.96014 |
acc_occpair | 0.87112 |
mIoU | 0.81346 |
pAcc | 0.87744 |
Find more results in visualizations.
This repo borrows heavily from deocclusion.