Liangfu Chen
Liangfu Chen
I have trained a ssd model with the features. Even the input image size is as large as 512x1024, the pretrained model works fine.
i my observation, there is no need to fix the parameters in mobilenetv2 while training ssd.
can you provide more detail on how did u train your model? and what exactly do you mean by referring to the limit of the model?
i have not tried the pretrained gluon model, but based on my previous experience with gluon models, i think there is a need to rename the model names, due to...
that make sense. i can upload pretrained model with multiplier with 0.75 and 0.5 then, so that more people could benefit from this.
I think mobilenetv2 based ssdlite should be reproducible if you stick on the details described in the paper. Can you tell which layers did you feed into the detection layers?...
I mean to reproduce the mAP results stated in the paper. IMHO, "the last layer" you highlighted means the shortcut layers in mobilenetv2, not the batchnorm layer.
The comment only shows size of input and output of an inverted residual block, and the tensor sizes that has been used internally inside the block are not illustrated in...
I think the paper I got on my hand is the version that is published in January, there is a small mistake in the illustration of stride size. But when...
I'm glad you would try the inference code, however, it requires an PR submission [here](https://github.com/dmlc/nnvm/pull/435) . For a quick solution, you can simply add `'clip'`to the `_identity_list` variable in `frontend/mxnet.py`...