HaohuaLv/retina-backbone_resnet50-ft_widerface

Retinaface Face Detection

A retinaface model for Face Detection trained on widerface dataset.

Notice: This is face detection model's training, evaluation and inference scripts in HuggingFace🤗 style from scratch for practice.

Train

Run

python train.py --model_config_file <MODEL_CONFIG_FILE>

can be found in folder config. Model checkpoints will be saved in folder checkpoints by default.

backbone-ResNet50 checkpoint can be download in my Google Drive or HuggingFace🤗.

Inference

Observe logits map and predicted bboxes

Run

python inference.py --checkpoint_path <CHECKPOINT_PATH>

is a model folder containing config.json and pytorch_model.bin.

Detect

Run

python detect.py --checkpoint_path <CHECKPOINT_PATH> --image_path <IMAGE_PATH> --save_path <SAVE_PATH>

References

• Retinface-pytorch