Prerequisites

1. Install tensorflow-gpu by running conda install -c anaconda tensorflow-gpu
2. Install NVIDIA dependencies for using GPU https://gist.github.com/msis/108a74d08f55eed48d8521fa968851ea

Data Preparation

1. Clone the repo

git clone https://github.com/experiencor/basic-yolo-keras

2. Make a dataset folder and organize your dataset within it into 4 folders:

• train_image_folder
• train_annot_folder : contains labels in VOC format (.xml)
• valid_image_folder
• valid_annot_folder : contains labels in VOC format (.xml)

3. Edit the following fields within config.json:

• "labels": put all the labels that exist in your dataset as a list (e.g. ["boat", "person", "bird"])
• add the path to your dataset under "train/valid_image_folder" and "train/valid_annot_folder"
• "saved_weights_name": give a nice name
• (Optional) "architecture": select from tiny yolo, full yolo, squeezenet, mobilenet, and inceptionV3

4. Download pretrained weights for backend (tiny yolo, full yolo, squeezenet, mobilenet, and inceptionV3) at: https://1drv.ms/f/s!ApLdDEW3ut5fec2OzK4S4RpT-SU and put the weights in the project's root. Keep the "pretrained_weights" field within config.json EMPTY.

5. Generate anchors for your dataset:

python gen_anchors.py -c config.json

Copy the generated anchors printed on the terminal to the "anchors" setting in config.json.

Training

1. Warm up the network

• Set warmup_epochs in config.json to 3.
• python train.py -c config.json

This process saves the trained weights to the file specified in saved_weights_name setting.

[Note: if there is a ResourceExhaustedError or MemoryError, try decreasing the "batch_size"]

2. Actual network training

• Set pretrained_weights setting in config.json to the warmup weights (whatever in saved_weights_name).
• Set warmup_epochs in config.json to 0.
• python train.py -c config.json

By the end of this process, the code will write the weights of the best model to file best_weights.h5 (or whatever name specified in the setting "saved_weights_name" in the config.json file). The training process stops when the loss on the validation set is not improved in 3 consecutive epochs.

Test

python predict.py -c config.json -w /path/to/best_weights.h5 -i /path/to/image/or/video

It carries out detection on the image and write the image with detected bounding boxes to the same folder.