From the comparison of different models [https://github.com/AlexeyAB/darknet/issues/3114], I understand that adding LSTM on top of PAN does not increase the mAP a lot. Is that the answer to my question?
Yes, just about +10 - +20 AP50 for the bad dataset, and may be twice more for the good dataset.
Thanks for the fast response. What is your guess for "yolo_v3_spp_pan_lstm" 's mAP, compared to "yolo_v3_tiny_pan3"? I guess there won't be much improvement.
Also, do you know why yolov4 does not inherit LSTMs? Is it due to PAN?
Hi @utkutpcgl Could you send me the yolo_v3_spp_pan_lstm.cfg file? I haven't found anywhere
Hi @utkutpcgl Could you send me the yolo_v3_spp_pan_lstm.cfg file? I haven't found anywhere
Hi @utkutpcgl Could you send me the yolo_v3_spp_pan_lstm.cfg file? I haven't found anywhere
It was not available at the time I sent the message. I just was curious about it. Sorry...
Most helpful comment
Yes, just about
+10-+20AP50 for the bad dataset, and may be twice more for the good dataset.