Open a279780399 opened 9 months ago
If there are samples in your dataset that have frames fewer than 13, this situation may occur. You can replace the 284-th line in train.py by "pid_start = random.sample(list(range(1, volume-4)), max(int(volume * self.prob), 1))"
started training, this error will be reported. Train step sometimes reports errors of over 2000, and sometimes errors of several hundred。