Day55 - P234+CBAM改进在VisDrone2019上的表现

yolov5s/m和yolov8s,删除P5检测头+添加P2检测头,在backbone的C3/C2f模块后添加一层CBAM,在VisDrone2019上训练200epochs

• Yolov5s-p234-cbam

  Parameters: 8.626M

  Class	Precision	Recall	mAP50	mAP50-95
  all	0.526	0.423	0.436	0.266
  pedestrian	0.572	0.48	0.514	0.248
  people	0.57	0.376	0.414	0.17
  bicycle	0.298	0.19	0.164	0.0778
  car	0.721	0.823	0.839	0.603
  van	0.536	0.478	0.488	0.347
  truck	0.507	0.351	0.368	0.252
  tricycle	0.45	0.298	0.307	0.176
  awning-tricycle	0.327	0.194	0.163	0.108
  bus	0.705	0.542	0.589	0.441
  motor	0.57	0.499	0.51	0.239

• Yolov5m-p234-cbam

  Parameters: 23.546M

  Class	Precision	Recall	mAP50	mAP50-95
  all	0.562	0.449	0.467	0.286
  pedestrian	0.615	0.493	0.542	0.266
  people	0.593	0.394	0.438	0.185
  bicycle	0.336	0.214	0.19	0.0897
  car	0.764	0.824	0.85	0.619
  van	0.566	0.483	0.505	0.368
  truck	0.603	0.407	0.426	0.293
  tricycle	0.498	0.35	0.348	0.201
  awning-tricycle	0.338	0.211	0.192	0.121
  bus	0.707	0.582	0.628	0.457
  motor	0.595	0.531	0.547	0.264

• Yolov8s-p234-cbam

  Parameters: 10.564M

  Class	Precision	Recall	mAP50	mAP50-95
  all	0.543	0.429	0.446	0.272
  pedestrian	0.597	0.474	0.521	0.252
  people	0.577	0.384	0.426	0.179
  bicycle	0.305	0.204	0.169	0.0792
  car	0.747	0.819	0.842	0.611
  van	0.573	0.474	0.493	0.356
  truck	0.548	0.363	0.378	0.256
  tricycle	0.439	0.315	0.305	0.177
  awning-tricycle	0.373	0.207	0.184	0.121
  bus	0.682	0.55	0.617	0.444
  motor	0.587	0.502	0.524	0.248