Poor accuracy on HRSC2016 dataset

[TerrafYassin]

Hi sir,
When execute your model on HRSC2016 on data you given to me i got the following results :

Need i to crop sub images from original images with an overlap like DOTA dataset or there is another problem ?

[Shank2358]

Hi sir, When execute your model on HRSC2016 on data you given to me i got the following results :

Need i to crop sub images from original images with an overlap like DOTA dataset or there is another problem ?

Hi, Did you use the pre-trained weights? What is the epoch?

[TerrafYassin]

Yes sir , i used darknet53.weights and the training epochs stopped on epoch 130 , after tested the best model on Test dataset i got 0.54 as mAP.

[Shank2358]

Yes sir , i used darknet53.weights and the training epochs stopped on epoch 130 , after tested the best model on Test dataset i got 0.54 as mAP.

Hi, I'm re-experimenting this dataset with the settings you gave. I will tell you the result.

[Shank2358]

Yes sir , i used darknet53.weights and the training epochs stopped on epoch 130 , after tested the best model on Test dataset i got 0.54 as mAP.

Hi, these are my results, I have uploaded the weight to google drive.
https://drive.google.com/file/d/1-D3C7Ul07frv8Uz-wt_gT7uQOJEtoTTm/view?usp=sharing
I only trained it for 50 epochs to achieve mAP=95. I re-updated the code. Maybe you need to check that the converted data labels are correct.



This is the training log.
log.txt

I will reply to your other questions in the next few days.