plane segmentation

[ZhouXiner]

Hi, I try to use the plane segmentation in the ADVIO dataset you provides, however it cant perforn like the mask you provide after CRF interface.
I use the pretrained model you provide. and the raw image has resized already. But this stiil some hard code in interface.py like plane_num.
I dont know what I did wrong, could pls help me?

[sudarshan-s-harithas]

Hi,
I hope you have followed the instructions as given in the README, please give us some more information and context as to how you executed/modified the programs.
The CRF inference improves the output detection of the plane-segmentation model and the num_plane is the maximum number of planes that can be detected by the model in a single image, our model has been trained with num_plane =3 (if you would like to detect more planes re-train the model by changing this value). refer to the papers PlaneRecover and RPVIO for the theory.

Thank You

[ZhouXiner]

Thanks for reply, I think I follow the steps in README.

1. extract the raw images from bag and resize it to (320,192)
2. all the image names are genrated in test.txt forrunning inference.py.
   The test.txt are sth like
   0.516478.jpg
   0.533146.jpg
   ....
   3.I run the inference.py like this
   python inference.py --dataset={0} --output_dir={1} --test_list={2} --use_preprocessed=False".format(path,output,test_list)
   path: the resized image folder
   test_list: the test.txt mentioned in 2
   The context I changed in inference.py is
   a. change the test image name loading to suit the test.txt: test_files_list.append(FLAGS.dataset_dir + '/' + t_split[0] )
   b. change the ckpt_file default value to ~/rp-vio-ws/src/rp-vio/plane_segmentation/pretrained_model/model since the former model isn't exists
3. run the crf_inference.py and resize the output image to raw size.
   Thanks for answering again!

[sudarshan-s-harithas]

the procedure seems right, can you please share a few of the results that you have obtained?

[ZhouXiner]

https://1drv.ms/u/s!AgIBb31yefjshBkdTyttCYSzMy5-?e=sNQ7We

This is some masks I generate, Thanks

[sudarshan-s-harithas]

Are these detections the output of the CRF? or the Plane segmentation model

[ZhouXiner]

Sorry for delay, a busy day...
The results are sementation model, the CRF results is just similiar like this. For example the images of full floor are just segmented partly.
I will send you the related CRF results when I use my PC. Forgive me.
Thank you very much.

[ZhouXiner]

https://1drv.ms/u/s!AgIBb31yefjshEG4L2eGruYLfo4v?e=3ingrZ
This is the more detailed masks. Thanks a lot