search

gd2016229035 / faster-rcnn-rgbd

Two stream Faster-RCNN evaluated on NYU Depth V2 dataset for RGBD object detection task.
Other
19 stars 5 forks source link

can you share the formated nyu-v2 ? #3

Open 7-qiliu opened 5 years ago

gd2016229035 commented 5 years ago

@7-qiliu Sorry I have deleted this dataset... You can simply use "toolbox" from NYUv2 website(https://cs.nyu.edu/~silberman/datasets/nyu_depth_v2.html) or some other script to extract every "instance" in every image from "nyu_depth_v2_labeled.mat". And then find the minimum enclosing Rectangle for every instance. So now you can get bounding box label~ Then you can transfer them to PASCAL VOC format to run this code directly. Please note that the original faster-rcnn implementation(from https://github.com/chenyuntc/simple-faster-rcnn-pytorch ,PyTorch0.3 version) requires that every xml file should have its bbox label. NYU object detection dataset Maybe has empty label , so you can simply delete them~

7-qiliu commented 5 years ago

@gd2016229035 thank you. how to transfer to PASCAL VOC format?

gd2016229035 commented 5 years ago

@7-qiliu If you get bbox information(class_id, xmin, ymin, xmax, ymax), you can easily transfer to VOC format. Maybe you can refer to this scrip(https://github.com/maozezhong/CV_ToolBox/blob/master/KITTI_2_VOC/txt_to_xml.py) which transfer "txt" format to PASCAL VOC "xml".

7-qiliu commented 5 years ago

@gd2016229035 got it, thank you so much

tuan-aislab commented 4 years ago

@gd2016229035 got it, thank you so much

Could you share the formated one?