search

zhaoweicai / mscnn

Caffe implementation of our multi-scale object detection framework
404 stars 211 forks source link

Evaluation of deconvolution on KITTI test benchmark #35

Open hekim34 opened 7 years ago

hekim34 commented 7 years ago

I have a question about the performance of deconvolution applied in ms-cnn. did you check the performance of object detection with deconvolution on KITTI benchmark test images? I want to know how much it affected your detection performance on KITTI benchmark test images. (not on validation set) I tried to use deconvolution features in other deep learning based detection algorithms(ex) SSD: Single Shot MultiBox Detector by Wei Liu), but didn't get the expected performance improvement.

zhaoweicai commented 7 years ago

I didn't extensively evaluate on test set, because it is not allowed. I only have two submissions: mscnn-8s-768-trainval and mscnn-7s-576-2x-trainval. The later has even better performance than the former one for car, which confirms that deconvolution helps to some extent. I also have some preliminary results on COCO, which also show deconvolution helps.

But it doesn't mean it always helps. For example, if the input image is already big enough, there is no need to use deconvolution, as explained in my paper. At least, deconvolution won't loss any information and won't impair the performance.