saketkunwar / cvwc2019_det

Tiger detection using yolov3 and yolov3-tiny open_vino inference or pytorch

cvcw2019_det

Tiger detection using yolov3 and yolov3-tiny on intel open_vino inference or pytorch

Inference Openvino usage Install openvino (https://docs.openvinotoolkit.org/latest/_docs_install_guides_installing_openvino_linux.html)

and setup environment paths for openvino

cd OpenVINO-YoloV3/

For yolov3 tiny-> python3 openvino_tiny-yolov3_tiger_submit.py --model ../openvino_converted_model/km_adam_yolov3_tinytiger_ckpt_9.xml --test_dir dir-to-test --result result-file

For yolov3-> python3 openvino_yolov3_tiger_submit.py --model ../openvino_converted_model/n_yolov3_tiger_13.xml --test_dir dir-to-test --result result-file

---

Inference pytorch usage

./test_tiger.sh path-to_test-dir result-filename

./test_tiger_tiny.sh path-to-test-dir result-filename

---

Train Pytorch only(https://github.com/eriklindernoren/PyTorch-YOLOv3) cd PyTorch-YOLOv3 cd weights ./download_weights.sh Required for imagenet pretrained weights(i.e yolov3.weights and yolov3-tiny.weights)

To train yolov3

./train_tiger

To train yolov3 tiny

./train_tiger tiny

---

Conversion of model from pytorch to tensorflow please visit https://github.com/PINTO0309/OpenVINO-YoloV3 and follow the guidelines ,specifically convert_weights_pb.py. Converion of model from tensorflow to openvino Follow (https://docs.openvinotoolkit.org/latest/_docs_MO_DG_prepare_model_convert_model_Convert_Model_From_TensorFlow.html)

Data: Train detection images needs to be downloaded from (https://cvwc2019.github.io/challenge.html) and stored in PyTorch-YOLOv3/data/tiger ( create images folder and store training images there). Orginal annotatino data in pascal-voc format was convereted and is already present in the repo.