Open Disciple7 opened 7 months ago
And when using tools/analysis_tools/get_flops.py to calculate FLOPs, I got about 100 GFLOPs with input shape=(1333, 800), 50 GFLOPs less than the result in the paper. Baseline also slightly less than 500 GFLOPs. I guess there are some unsupport ops in the model. How to get the correct answer ?
当使用tools/analysis_tools/get_flops.py计算FLOP时,我得到了大约100个GFLOP,输入形状=(1333, 800),比本文的结果少了50 GFLOPs。基线也略低于500 GFLOP。我想模型中有一些不支持的操作。如何得到正确的答案?
hi, I didn't find instances_UAVtrain_v1.json in the Visdrone dataset, where should I find it
I noticed that the paper mentioned the GFLOPs of the model, which was used to prove the effect of spconv. But how to get it ? Did the author calculate this by hand ? Or is there any useful code to get it automatically ?
我注意到论文中提到了该模型的GFLOPs,用于证明spconv的效果。但是如何获得它呢?这是作者亲手计算的吗?或者有什么有用的代码可以自动获取它吗?