Skip to content

Latest commit

 

History

History
63 lines (53 loc) · 1.99 KB

GETTING_STARTED.md

File metadata and controls

63 lines (53 loc) · 1.99 KB
Raw

Data Preparation

COCO dataset

  1. Download and extract COCO 2017 train and val images with annotations from http://cocodataset.org.
  2. Download panoptic annotations from COCO website.
  3. Download the pre-processing panopitc segmentation annotations from here(Google Drive).
  4. prepare the data as the following structure:
detectron2/
  datasets/
    coco/
      {train,val}2017/
      panoptic_{train,val}2017/  # png annotations
      annotations/
        panoptic_{train,val}2017.json
        panoptic_{train,val}2017_trans/  # pre-processing panoptic segmentation png annotations

ADE20K dataset

  1. Download and extract the ADE20K dataset train and val images from http://sceneparsing.csail.mit.edu/.
  2. Download the annotations for panoptic segmentation from here (Google Drive).
  3. prepare the data as the following structure:
detectron2/
  datasets/
    ADE20K_2017/
      images/
        training/
        validation/
      new_segment_anno_continuous/
        training/
        validation/
      ade_{train,val}_things_only.json
      panoptic_ade20k_val_iscrowd.json

Getting Started

Training & Evaluation in Command Line

To train a model, first cd detectron2 and then run

python tools/train_net.py --num-gpus 4 \
  --config-file ./configs/ADE/panoptic_fpn_bs8_R_50_1x_G.yaml

To eval a model, use

./train_net.py \
  --config-file  ./configs/ADE/panoptic_fpn_bs8_R_50_1x_G.yaml \
  --eval-only MODEL.WEIGHTS /path/to/model_weights

Models

Model weights

Datasets Google Drive
ADE20K Download link
COCO Download link