Skip to content

Latest commit

 

History

History
45 lines (30 loc) · 1.43 KB

GETTING_STARTED.md

File metadata and controls

45 lines (30 loc) · 1.43 KB

Getting Started with OneFormer

This document provides a brief intro of the usage of OneFormer.

Please see Getting Started with Detectron2 for full usage.

Training

  • Make sure to setup wandb before training a model.

    pip install wandb
    wandb login
  • We provide a script train_net.py, that is made to train all the configs provided in OneFormer.

  • To train a model with "train_net.py", first setup the corresponding datasets following datasets/README.md.

  • Be default, the model uses task=panoptic for evaluation during training.

python train_net.py --dist-url 'tcp://127.0.0.1:50163' \
    --num-gpus 8 \
    --config-file configs/ade20k/swin/oneformer_swin_large_bs16_160k.yaml \
    OUTPUT_DIR outputs/ade20k_swin_large WANDB.NAME ade20k_swin_large

Evaluation

  • You need to pass the value of task token. task belongs to [panoptic, semantic, instance].

  • To evaluate a model's performance, use:

python train_net.py --dist-url 'tcp://127.0.0.1:50164' \
    --num-gpus 8 \
    --config-file configs/ade20k/swin/oneformer_swin_large_bs16_160k.yaml \
    --eval-only MODEL.IS_TRAIN False MODEL.WEIGHTS <path-to-checkpoint> \
    MODEL.TEST.TASK <task>

Inference Demo

We provide a demo script for inference on images. For more information, please see demo/README.md.