Open lxy5513 opened 11 months ago
@airockchip 您好，请教一个问题，为何我转化的时候cos_similarity很高，但是onnx和rknn模型的mAP差距很大呢? (我使用的是yolov6)
---> time_info: model_init: 2239.471924 ms input_set: 21.114054 ms run: 624.894653 ms output_get: 1.16285 ms ---> compute onnx result For example iter 1/1 rknn(fp) VS onnx cos_similarity: output-0 : 0.9999998807907104 output-1 : 0.9999695420265198 output-2 : 0.9999690055847168 output-3 : 1.0000001192092896 output-4 : 0.9999822378158569 output-5 : 0.9999829530715942 output-6 : 1.0 output-7 : 0.9999997615814209 output-8 : 0.9999997019767761
onnx metrics: map --> 0.7929503520008853 map50--> 0.9390134759379819 map75--> 0.8890040657875151 map85--> 0.9143185738930419 map95--> 0.7839896214896215
rknn metrics: map --> 0.19587591047841849 map50--> 0.28743282298483197 map75--> 0.2165340249311251 map85--> 0.3492236917768833 map95--> 0.29946581196581196
兄弟你解决了吗，我也是这样
Calsia
commented
2 months ago Calsia
commented
2 months ago
@airockchip 您好，请教一个问题，为何我转化的时候cos_similarity很高，但是onnx和rknn模型的mAP差距很大呢? (我使用的是yolov6)
---> time_info: model_init: 2239.471924 ms input_set: 21.114054 ms run: 624.894653 ms output_get: 1.16285 ms ---> compute onnx result For example iter 1/1 rknn(fp) VS onnx cos_similarity: output-0 : 0.9999998807907104 output-1 : 0.9999695420265198 output-2 : 0.9999690055847168 output-3 : 1.0000001192092896 output-4 : 0.9999822378158569 output-5 : 0.9999829530715942 output-6 : 1.0 output-7 : 0.9999997615814209 output-8 : 0.9999997019767761
onnx metrics: map --> 0.7929503520008853 map50--> 0.9390134759379819 map75--> 0.8890040657875151 map85--> 0.9143185738930419 map95--> 0.7839896214896215
rknn metrics: map --> 0.19587591047841849 map50--> 0.28743282298483197 map75--> 0.2165340249311251 map85--> 0.3492236917768833 map95--> 0.29946581196581196