search

ruotianluo / pytorch-faster-rcnn

pytorch1.0 updated. Support cpu test and demo. (Use detectron2, it's a masterpiece)
MIT License
1.82k stars 475 forks source link

Training data consists of oriented bounding boxes rather than horizontal ones #93

Open vsd550 opened 6 years ago

vsd550 commented 6 years ago

I want to perform classification and detection only on my dataset of satellite images using Faster Rcnn. The problem is that there are very dense objects in the image and have oriented bounding boxes with all the 4 coordinates specified, while the faster rcnn takes horizantal bounding boxes with 2 coordinates only(Xmin,Ymin,Xmax,Ymax) while training. SO basically we need to regress over 4 coordinates rather than 2. Please suggest how can I do it or refer me to some similar projects. Thanks in advance

Shamsa279 commented 6 years ago

@vsd550 did you find something related to your problem? Actually I am working on the same problem

vsd550 commented 6 years ago

@Shamsa279 Unfortunately not yet. I am on it. Please share if u find something. Thanks

priya55612 commented 4 years ago

@Shamsa279 and @vsd550 Did you guys find anything on this?

priya55612 commented 4 years ago

My bounding boxes have center (x,y), height, width, and angle