Skip to content

Latest commit

 

History

History
executable file
·
51 lines (43 loc) · 1.69 KB

GETTING_STARTED.md

File metadata and controls

executable file
·
51 lines (43 loc) · 1.69 KB
Raw

Getting Started with Panoptic-DeepLab

Simple demo

Please download pre-trained model from MODEL_ZOO, replace CONFIG_FILE with corresponding config file of model you download, and then run

python tools/demo.py --cfg configs/CONFIG_FILE \
    --input-files PATH_TO_INPUT_FILES \
    --output-dir PATH_TO_OUTPUT_DIR \
    TEST.MODEL_FILE YOUR_DOWNLOAD_MODEL_FILE

If you want to merge image with prediction, you can add a --merge-image flag:

python tools/demo.py --cfg configs/CONFIG_FILE \
    --input-files PATH_TO_INPUT_FILES \
    --output-dir PATH_TO_OUTPUT_DIR \
    --merge-image \
    TEST.MODEL_FILE YOUR_DOWNLOAD_MODEL_FILE

Training & Evaluation in Command Line

We provide a script in "tools/train_net.py", that is made to train all the configs provided in panoptic-deeplab.

To train a model with "train_net.py", first setup the corresponding datasets following datasets/README.md, then run the following command with NUM_GPUS gpus:

python -m torch.distributed.launch --nproc_per_node=NUM_GPUs tools/train_net.py --cfg configs/CONFIG_FILE

To train a model with a single GPU:

python tools/train_net.py --cfg configs/CONFIG_FILE TRAIN.IMS_PER_BATCH 1 GPUS '(0, )'

To evaluate a model's performance, use

python tools/test_net_single_core.py --cfg configs/CONFIG_FILE

To evaluate a model with test time augmentation (e.g. flip and multi-scale), use

python tools/test_net_single_core.py --cfg configs/CONFIG_FILE \
    TEST.TEST_TIME_AUGMENTATION True \
    TEST.FLIP_TEST True \
    TEST.SCALE_LIST '[0.5, 0.75, 1.0, 1.25, 1.5, 1.75, 2]'