hello, I use the code to finetune the dataset of DETRAC, which is also a dataset for car detection, but after I train the network for 50000 times, the detector can't detect any objects, the same situation for the model after 5000 times training. Before training, I have modified the size of input to 1024x768 to avoid out of memory, and changed the learn rate from 0.0005 to 0.00001 to avoid the loss changing to NAN. So please help me why the trained detector didn't work? Thank you very much.
jyuan1118
commented
6 years ago
hello, I use the code to finetune the dataset of DETRAC, which is also a dataset for car detection, but after I train the network for 50000 times, the detector can't detect any objects, the same situation for the model after 5000 times training. Before training, I have modified the size of input to 1024x768 to avoid out of memory, and changed the learn rate from 0.0005 to 0.00001 to avoid the loss changing to NAN. So please help me why the trained detector didn't work? Thank you very much.