Guanghan / ROLO

ROLO is short for Recurrent YOLO, aimed at simultaneous object detection and tracking
Apache License 2.0
883 stars 278 forks source link

When training, why it requires the pre-trained model to be put in the output directory? #21

Open hanweiWill opened 7 years ago

hanweiWill commented 7 years ago

I attempted to run the training experiments of ROLO, Taking experiments/training/ROLO_step1_train_30_exp2.py for example: after I set the rolo_weights_file = 'output/ROLO_model/model_step1_exp2.ckpt in line 63 to a local directory that I created, the program kept returns the error saying Restore called with invalid save path. I tried downloaded the model_step1_exp2.ckpt file from the given link and put it in the directory. Then the training works fine. So my question is why the training process require a pre-trained model to exist in the output directory?