Gumpest / YOLOv5-Multibackbone-Compression

YOLOv5 Series Multi-backbone(TPH-YOLOv5, Ghostnet, ShuffleNetv2, Mobilenetv3Small, EfficientNetLite, PP-LCNet, SwinTransformer YOLO), Module(CBAM, DCN), Pruning (EagleEye, Network Slimming), Quantization (MQBench) and Deployment (TensorRT, ncnn) Compression Tool Box.
996 stars 202 forks source link

CBAM注意力模块问题 #12

Closed xiaoyi-creator closed 6 months ago

xiaoyi-creator commented 2 years ago

您好,我想问一下,在加了第四层预测头部之后,您是在head的前三个C3之后增加了CBAM模块,但是为什么在预测的时候我们的预测输出是C3之后就进行预测了呢,不应该在CBAM模块之后进行预测吗? [[23, 27, 31, 35], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5)`,是不是 应该改为 [[24, 28, 32, 35], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5)`,

Gumpest commented 2 years ago

您好,这是原作者的意图;按照论文结构应该是您说的那样 [[24, 28, 32, 35], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5)

xiaoyi-creator commented 2 years ago

奥奥,好的,还有一个问题想问您一下,就是原论文中的WBF代替NMS,我在您的代码中没有看到,想请问一下是您对这部分内容没有复现吗

xiaoyi-creator commented 2 years ago

您好,我在您的yolov5xYPH.yaml文件中看到你在预测头的c3模块之前加了SPP模块,想问一下这里加SPP模块是为了更好的提取特征吗?

Sa-UpWorld commented 2 years ago

作者您好,我在yolov5xP2CBAM-Swin-BiFPN-SPP.yaml中看到您在C3模块前加了SPP模块,请问为什么要加池化部分呢?是为了同一输入尺寸吗?