search

hitachi-rd-cv / MILA

Memory-Based Instance-Level Adaptation for Cross-Domain Object Detection
14 stars 1 forks source link

about memory_feat_comic.pkl #3

Open Hyouka-qitanta opened 2 months ago

Hyouka-qitanta commented 2 months ago

when i train the model i got the follow error:

Traceback (most recent call last): File "train_net_mem.py", line 77, in launch( File "/home/zhengguida/miniconda3/envs/MILA/lib/python3.8/site-packages/detectron2/engine/launch.py", line 82, in launch main_func(*args) File "train_net_mem.py", line 70, in main return trainer.train() File "/home/zhengguida/MILA/memory/engine/trainer_mem.py", line 205, in train self.train_loop(self.start_iter, self.max_iter) File "/home/zhengguida/MILA/memory/engine/trainer_mem.py", line 224, in train_loop self.run_step_full_semisup() File "/home/zhengguida/MILA/memory/engine/trainer_mem.py", line 385, in run_step_full_semisup file = open("memory_feat_comic.pkl",'rb') FileNotFoundError: [Errno 2] No such file or directory: 'memory_feat_comic.pkl'

how to get the memory_feat_comic.pkl file?

onkarkris commented 2 months ago

Sorry for the confusion. We don't need any pkl file to run this code; that was only for debugging purpose. I have updated trainer_mem.py; you should be able to run it without any issues now!

Hyouka-qitanta commented 2 months ago

Sorry for the confusion. We don't need any pkl file to run this code; that was only for debugging purpose. I have updated trainer_mem.py; you should be able to run it without any issues now!

Thank you for your prompt reply, i run the code successfully. I found the MAX_ITER in faster_rcnn_R101_cross_comic_08032.yaml file is 200000, it's too much and need almost 15 days to train on A100 GPU. I want to know does it really need to train 200000 iterations or just a mistake?