Skip to content

Latest commit

 

History

History
64 lines (43 loc) · 1.48 KB

03_Quickstart.md

File metadata and controls

64 lines (43 loc) · 1.48 KB

Getting Started with DotaTron

This section is a overview of initiating the detectron2 training and evaluation on oriented rotated datasets challenges. The primary usecase tested in the project is the DOTA-v2 dataset challenge. Refer to the README.md to prereq and review the DATASET_Details.md for data preparation and annotation format details.

Training

Train detection model

to train run: TODO

Makesure these model configs are correct:

DATA_LOADER.NUM_WORKERS 0 
NUM_GPUS 2 
TRAIN.BATCH_SIZE 16

Resume training from checkpoint

For training that has already started and wanting to retrain an existing model run: TODO

Tensorboard

cd $PATH_MODEL_LOG/outputs
tensorboard --logdir=.

Test and Eval

Visualize Detection Results

To test the model inference run the following or run sh scripts/test_train.sh

INPUT_DIR=
OUTPUT_DIR=
CONF=
USE_GPU=
CKPT_PATH=
NUM_IMAGES=20
python test_tron.py --input_dir ${INPUT_DIR} --output_dir ${OUTPUT_DIR} --conf ${CONF} \
                    --use_gpu ${USE_GPU} --ckpt_path ${CKPT_PATH} --num_images ${NUM_IMAGES}

Evaluation

To test the model inference run the following or run sh scripts/test_train.sh


INPUT_DIR=
OUTPUT_DIR=
CONF=
USE_GPU=True
CKPT_PATH=
NUM_IMAGES=20
python test_tron.py --input_dir ${INPUT_DIR} --output_dir ${OUTPUT_DIR} --conf ${CONF} \
                    --use_gpu ${USE_GPU} --ckpt_path ${CKPT_PATH} --num_images ${NUM_IMAGES}