magicboomliu / orientation-learning-for-autonomous-driving

Simple Orientation Learning using ResNets

OrientationLearning

Simple Orientation Learning using ResNets

• Download the VehicleOrientationDataset
  https://github.com/sekilab/VehicleOrientationDataset

Note that in this repo, we using consider the category of 'car', which means 3 categories:

• car_back : 0
• car_side: 1
• car_front :2

for the training, we use vehicle-orientation 1~4 for training(almost 20K images), and use vehicle-orientation-5 for evaluation.

For trainining, we use:

cd scripts 
sh train.sh

For inference on new images, we provided the inference code with pre-trained model:

• Download the pretrained model

sh download_model.sh

The Downloaded model with be in pretrained_models/model_best.pth

Then run the inference code:

python inference.py --pretrained_model_path "pretrained_models/model_best.pth" --nums_layers 34 \
> --image_path "<YOU IMAGE PATH>" --annotation_path "<ANNOTATION JSON FILE>" --saved_folder "<SAVED FOLDER >"