woolpeeker
woolpeeker
There is no OOM error reported. The system is Ubuntu18.04 and the evaluation is running in a docker container. The max RAM of my computer is 64G, GPU is 2080Ti-12G...
I change the eval cmd to ```ROUTES=assets/routes_lav_valid.xml ./leaderboard/scripts/run_evaluation.sh``` it success to finish the route0 test, which cost about 1 hour The memory usage is also huge, reaching max 37G. If...
I'm also curious about it. Have you figured it out yet? Should the pre-trained model output the exact same result as the paper? I got totally different results.
It seems the same question with https://github.com/Diego999/pyGAT/issues/4#issue-343470929
I change the data split to the official GAT style, the accuracy on cora can only reach 0.817. Is there any solution to improve the performance
https://github.com/hhaAndroid/mmdetection-mini/blob/3858c8c2f071c064fe78ce24088a1c9815ae1a21/mmdet/models/dense_heads/rr_yolov5_head.py#L73 I think the reason is that the yolov5Head‘s output channel number is hard coded.