Linus comments

Results 24 comments of


                                            Linus

Problem when try to run inference

@triangleCZH --load_detectron is for loading pretrained detectron in original caffe model(pkl or pth) but in this case original caffe model's weight name(layer name) is different with pytorch version's layer name....

Loss NaN about using vovnet as backbone in RetinaNet

When I tested this kind of lightweight backbone in object detection (ex, mobilenet, shufflenet etc..), i set warm up iter longer.

error: expected template-name before ‘<’ token class _NewEmptyTensorOp : public torch::autograd::Function<_NewEmptyTensorOp>

what is your libtorch version? I think something changed in new libtorch.

How do you convert a python trained model into c++

My converting script is very naive. It just extracts weights from backbone and heads seperately. I am not sure if i understood 'the inference will affect the result of inference',...

How do you convert a python trained model into c++

This project is not as fast as you expect and need more optimization. I don't recommend it, if you need hight performance, speed and memory efficient.

How do you convert a python trained model into c++

So, you mean that faster rcnn's process must be under 1 second per many images?

How do you convert a python trained model into c++

How much images per second? It really depends on the number of images, backbone type, the number of classes and nms size. If you can accpet the drop of accuracy,...

How do you convert a python trained model into c++

I think you need ablation study using nms. NMS is dominant factor in speed.

How do you convert a python trained model into c++

check this config option PRE_NMS_TOP_N_TEST: 1000 POST_NMS_TOP_N_TEST: 1000 FPN_POST_NMS_TOP_N_TEST: 1000

How do you convert a python trained model into c++

Decreasing these options can potentially drop the accuracy. But not always, when i tested it to half NMS candidates, it only dropped 0.01 mAP. So, you have to adjust and...