Training image loading problems

[Chenhanne]

I use a custom coco dataset for training and there are instances where the training images are not fully loaded.
I have 2583 training pictures and 884 test pictures.
As shown, only half of the training images can be detected. But all the test images can be detected.

To find the problem, I used the training set as a test set and the test set as a training set to check if it was a problem with the data that caused the images to load incompletely. However, the results are as follows:

It is easy to see that the original training set was able to be fully loaded (2583 images) after being used as a test set. However, the original test set was only half recognised (442 images) after being used as a training set.
It seems that only the training process has the problem
What's going on here? How can I solve this problem?

[Chenhanne]

The model I use is:
base = [
'../base/models/faster-rcnn_r50_fpn.py',
'../base/datasets/coco_detection.py',
'../base/schedules/schedule_1x.py',
'../base/default_runtime.py'
]

[aso538]

According to your statement and the corresponding configuration file, if you haven't adjusted the batch size, then it should be that one batch in the training set contains two images

Just modify the value of batchsize

[Chenhanne]

Thank you so much for your answer!!! At 2023-12-06 15:41:23, "aso538" ***@***.***> wrote: According to your statement and the corresponding configuration file, if you haven't adjusted the batch size, then it should be that one batch in the training set contains two images image.png (view on web) Just modify the value of batchsize — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>