SSD Fast Transfer Learning

No folder is added for SSD fast transfer learning. Add extra command parameters to the original SSD basic detection code to enable SSD fast transfer learning. You are advised to perform the function in the ssd_mobilenet_fpn_mindspore folder.

Training Command Reference

Run the following command to start training:

python3 model_train.py --train_dataset_path={train_dataset_path} --train_output_path={train_output_path} --pretrained_ckpt_path={pretrained_ckpt_path} --epoch_size=100 --batch_size=4 --input_width=2048 --input_height=1536 --init_lr=0.001 --device_num=0 --run_eval=True --quick_train=True --eval_start_epoch=5

The training logs and file generation of SSD fast transfer learning are similar to those of SSD basic detection. For details, see Table 1.

python3 model_eval.py --eval_dataset_path={eval_dataset_path} --eval_ckpt_path=./output_dir --eval_output_path=./eval_result --min_score=0.01 --nms_threshold=0.2 --pretrained_ckpt_path= pre_trained_ckpt