search

Karel911 / TRACER

TRACER: Extreme Attention Guided Salient Object Tracing Network (AAAI 2022) implementation in PyTorch
Apache License 2.0
195 stars 41 forks source link

Runtime Speeds #10

Closed vjsrinivas closed 2 years ago

vjsrinivas commented 2 years ago

Hi, thanks for putting out an interesting paper and the code. I was testing it with the various pretrained weights on DUT-TE, and I'm getting an end-to-end speed of ~11 FPS on an RTX 2070 Super. Is that an expected speed?

Karel911 commented 2 years ago

Hi, Thank you for your interest in our research. Please check the batch size of 32 and exclude the lines related with measures. The measuring costs a few minutes so that TRACER shows lower FPS than the original ones.

vjsrinivas commented 2 years ago

Hi, The speeds listed was with batchsize of 1. I commented out the measures lines (the references to AvgMeter and the Eval_tool's cal_total_metrics and saving the output). Is there anything to comment out? I'm getting 15-16 FPS now.

Karel911 commented 2 years ago

I tested TE2 under the AMD-RYZEN R9 5900X CPU and single RTX 3090 GPU with batch size of 32. I obtained the same FPS results listed in our paper. The FPS inevitably depends on the hardware so that it results in the FPS discrepancy. Please set the batch size of 32 (as you can) and re-check the FPS.

Thanks,