Project: Machine Learning models for image classification, object detection and segmentation in Virtual Reality

Projects Radiology
, , ,
21

I was running experiments is Kaggle, but can create a GitHub repo to store them. Is it ok?

UPD: here is my preliminary repo: GitHub - DaniilOr/cv_in_vr: My GSoC 2022 project

I will add more details to README soon (project/company info, the purpose of this project, repository structure etc)

22

Please use Gitlab under the LibreHealth organization.
@pri2si17 or @r0bby can create one for you under the Radiology project.

23

Got it, I will migrate to gitlab. Thanks!

24

@DaniilOr

Please use following repository to commit your codes:

@sunbiz @judywawira Please open gitlab issues in this repository.

2 Likes
26

I updated the naming but otherwise, this is fine – usually, I do: lh-PROJECTNAME-REPONAME which is sane IMHO and not too burdensome. PROJECTNAME is radiology in this case. REPONAME is entirely up to you.

1 Like
27

My tasks:

  1. Move notebooks with output to gitlab
  2. Do one more multi class segmentation task

Main blockers:

  1. When downloading models from Kaggle, output can be vanished

My expectations / what help I expect to receive:

  1. Probably, I will need some feedback on my notebooks: approach, code style, information etc
28

Hi @sunbiz can @DaniilOr get GPU server to train the model? Please let us know.

29

@pri2si17 Hi, are you joining the meeting tonight?

30

My work for the past week:

Tasks for next week: 0. Add something to experiments if needed

  1. Commit codes for model inference (probably we have to discuss here the inference of the model)
  2. Implement method for distorting bboxes for object detection

Main blockers:

  1. distorted bounding-boxes may not work well (in this case other options such as the change of bboxes shape could help)
  2. It is not 100% clear how the inference should be implemented

My expectations / what help I expect to receive:

  1. Guidances regarding the inference
31

Hi @DaniilOr , sorry I got busy so I couldn’t join the meeting. I will have a look at the code and let you know.

1 Like
32

That is fine. Waiting for any kind of feedback

1 Like
33

My work for the past week:

  • Created an inference pipeline (link)

Tasks for next week:

  1. Add something to inference if needed
  2. Show findings on bbox transformation

My expectations / what help I expect to receive:

  1. Feedback on the inference
1 Like
34

Have you been blogging?

35

Sure, I was. Probably, some changes in the post were not saved, I will check it out

36

Just share it here weekly – it’s a simple post and should not take too much extra time :slight_smile:

37

I was blogging here https://daniilor.blogspot.com, so will just share links to new posts

38

Yeah, that’s exactly what I want. It’s an easy way for those reading this thread to follow what you’re doing

39

Hi @DaniilOr this link is not working. Also, can you please provide work updates on forum for this week. I will meet you next week.

1 Like
40

Sure. It is possible that the platform I am using requires VPN in some countries (idk why). Here is the link to RSS feed of my blog: feed:Daniil Orel GSoC 2022 blog

My work for the past week:

  • Experimented with representation of bounding boxes for object detection in VR environment.
  • Found out that when an object is centred, we may just leave an original bounding box. For peripheral objects small corrections are needed (right now, using ellipses used to tackle this issue, but I am planning to use point-distortion algorithm)

Tasks for next week:

  • Transfer Segmentation models to TensorRT / update their inference
  • Come up with the final algorithm for bounding boxes in VR

My expectations / what help I expect to receive:

  1. Feedback on MRs

PS good luck with your conference

41

Hi @DaniilOr Thanks for the update. I am talking about the gitlab link. Not Found (gitlab.com)

Your fork is not available to review the code. Please fix the url.