yehao

soft-nms cannot improve mAP used SSD detection(one-stage)

4

I implement soft-nms in SSD(caffe), code link [modify detection_output.cu, bbox_util.hpp, bbox_util.cpp](https://github.com/ujsyehao/hardcaffe), I ensure the custom implementation version is the same as author-provided python version(I have tested it use fake input)...

mobilenet v2-ssd?

2

Can you provide mobilenetv2-ssd model?

Hi, I use your code to do SSD-variant model conversion, It has the error: ` NvPluginSSD.cu:428: virtual void nvinfer1::plugin::PriorBox::configure(const nvinfer1::Dims*, int, const nvinfer1::Dims*, int, int): Assertion H == inputDims[0].d[1] failed.`...

When I use the demo to de video detection, It has the error: `runtime.cpp (16) - Cuda Error in allocate: 2 terminate called after throwing an instance of 'nvinfer1::CudaError' `...

Why need implement softmax plugin?

1

In tensorRT 4.0 doc: It supports softmax on C, H, W dimension

invalid syntax: f"conv_{module_i}"

5

When I run `python3 test.py --model_def config/prune_yolov3-hand.cfg --weights_path /home/yehao/Downloads/prune_yolov3_ckpt.pth --data_config config/oxfordhand.data --class_path data/oxfordhand.names --conf_thres 0.01`. It has the error: 

Paper says 'replace a 7x7 convolution with three 3x3 convolutions'. I am confused about it. ex: input: 224x224 channel=3 7x7 conv pad=3 stride=2 channel=64-> output dimension: 112x112 Its flops is:...