chenjshnn
commented
3 years ago Actually, different names mean different anchor box settings while training the model. They all use the same dataset, i.e., Rico.
These names correspond to Table 3.
Anchor box setting (size & aspect ratio) Faster RCNN - Rico Customized: 32, 64, 128, 256, 512, 1:1, 2:1, 4:1, 8:1 Faster RCNN - Rico Default: 128, 256, 512, 1:1, 2:1, 4:1, 8:1 yolov3 - rico5box: use 5 anchor boxes, i.e., k=5 yolov3 - rico: use 9 anchor boxes, i.e., k=9
您好，目前我在复现您论文中的一些数据，想请教下您Faster-RCNN和yolov3中的数据集不知道应该使用哪个。 Faster-RCNN中数据集有ricoCustomized、ricoDefault不太明白，yolov3中rico5box、rico也不太明白。您可以介绍下吗？