facebookresearch / maskrcnn-benchmark

Fast, modular reference implementation of Instance Segmentation and Object Detection algorithms in PyTorch.
MIT License
9.31k stars 2.49k forks source link

Why the validation loss is always increasing? #1357

Open imzhangyd opened 1 year ago

imzhangyd commented 1 year ago

❓ Questions and Help

The trainning loss and metrics seems normal during training, but the validation loss is increasing. Please see bolow information for details. I'm curious about the reason behind this issue, and whether there might be an error during the training process. If there is an error, could you please let me know how I can correct it?

Model, Dataset, Config

I trained mask-rcnn-R50-FPN-1x model on my own dataset. There is one object on each image in my dataset. I change some params: PRE_NMS_TOP_N_TRAIN: 200 PRE_NMS_TOP_N_TEST: 100 POST_NMS_TOP_N_TRAIN: 200 POST_NMS_TOP_N_TEST: 100 FPN_POST_NMS_TOP_N_TRAIN: 400 FPN_POST_NMS_TOP_N_TEST: 100 IMS_PER_BATCH: 8

Loss and metrics

The trainning loss seems normal, but the validation loss is always increasing. I don't know where's the problem.

The trainning loss: trainloss

The validation loss: validationloss

The metrics seems normal:

metrics_bbox metrics_segm