[Re] Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation

This is a downsampled re-implementation of Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation, trained on a single NVIDIA A10G. More details to follow.

Environment

To install requirements:

pip install -r requirements.txt

Training

To train the model detailed in the paper, run the following command:

TRAINING=TRUE/FALSE python -m src.model.deeplab

Project Structure

---

├── LICENSE
├── README.md               <- you are here!
├── report                  <- reproducibility challenge report
├── requirements.txt        <- training environment
└── src                     <- Source code for use in this project.
    ├── const.py
    ├── data
    │   ├── cityscapes.py   <- cityscapes dataloader creation
    │   └── common.py       <- Dataset-angosting preprocessing routines
    ├── model
    │   ├── aspp.py         <- atrous sparse pyramid pooling layer
    │   ├── decoder.py      <- semantic and instance decoders
    │   ├── deeplab.py      <- primary trainer
    │   ├── encoder.py      <- xception-71 backbone
    │   ├── heads.py        <- semantic, instance center and instance regression heads
    │   ├── loss.py         <- weighted bootsrapped cross-entropy for semantic head
    │   └── metrics.py      <- mIOU, AP, PQ
    └── coco_tools.py       <- pycocotools extensoions

---