search

sfzhang15 / RefineDet

Single-Shot Refinement Neural Network for Object Detection, CVPR, 2018
Other
1.43k stars 393 forks source link

multi_scale test..... #169

Closed AiIsBetter closed 5 years ago

AiIsBetter commented 5 years ago

Hi@sfzhang15 I didn't configure the environment for caffe, so I just looked at the code. In refine_test, input non-size pictures to net for detection, and then do some post-processing. But isn't a trained model that can only predict fixed-size images? So how does this work? Looking forward to your response.

sfzhang15 commented 5 years ago

@AiIsBetter RefineDet is a fully convolutional network, so it can has any size of input images. For single-cale testing, the testing input size should be the same as the training input size to have best performance. For multi-scale testing, we can use many scales to test the model and then merge these results.

AiIsBetter commented 5 years ago

@AiIsBetter RefineDet is a fully convolutional network, so it can has any size of input images. For single-cale testing, the testing input size should be the same as the training input size to have best performance. For multi-scale testing, we can use many scales to test the model and then merge these results.

I understand. I've tried it myself. That's it. Thank you.