I take the liberty to disturb you. I found that the result of the picture annotation is incorrect during the test

microsoft / RegionCLIP

[CVPR 2022] Official code for "RegionCLIP: Region-based Language-Image Pretraining"

Apache License 2.0

712 stars 52 forks source link

I take the liberty to disturb you. I found that the result of the picture annotation is incorrect during the test #52

Closed zmw1003 closed 1 year ago

zmw1003 commented 1 year ago

I strictly follow these two steps without any change 99b8e01c01095b4c0417418b12688a7 But the result of my output picture is like this, which makes me very distressed. I hope I can get help 3f74b1699e4fdc8fe298d40f36bd6a1 7fb486f8f6842fce20d4f0bdebe0b45

YiwuZhong commented 1 year ago

The scripts work well on other users' machines. It's not totally sure which part is not settled down on your side. You could try debugging the results. For example, in the visualization script, print the confidence scores of predicted boxes. The highest scores above are around 21% which is quite low.