Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

check / adjust behavior of v3 feature production for ROIs with no blob #29

Open
hsosik opened this issue Nov 20, 2017 · 5 comments
Open
Assignees

Comments

@hsosik
Copy link
Owner

hsosik commented Nov 20, 2017

In my effort to compare classification with v2 and v3 features for a large training set from MVCO, I found what I think is an inconsistent behavior in cases for ROIs with no blob. I think v2 features have an entry for those ROIs with 0s for any feature that requires a blob, but computed values for features that are based on only the raw image. It looks like the test set of v3 features Joe produced have those ROIs skipped altogether. I found this because the change in blob algorithm apparently occasionally produces no blob when previously there was one. Here is an example where there is an old blob (and corresponding v2 features, but no new v3 features:
http://ifcb-data.whoi.edu/mvco/D20170423T045058_IFCB010_02289.html
\sosiknas1\IFCB_dev\features_v3_test\features\D20170423T045058_IFCB010_fea_v3.csv

@joefutrelle
Copy link
Collaborator

I think you're just referring to HOG features, right? And the desired fix is to output 0s except for those?

@joefutrelle
Copy link
Collaborator

joefutrelle commented Nov 20, 2017

Reviewing the code, it looks like it's already designed to produce output with zeros except for image-only features, so there must be a bug in it somewhere.

https://github.com/hsosik/ifcb-analysis/blob/features_v3/python/ifcb_features/all.py#L461

@joefutrelle
Copy link
Collaborator

I tried two things:

  1. computing features for a synthetic noisy flat gray image. produced no blob, and correct feature output.
  2. computing features for the image linked to. produced a blob and correct feature output.

this was some sort of bookkeeping error writing out the CSV when I was batch processing. how many ROIs did this happen for?

@hsosik
Copy link
Owner Author

hsosik commented Nov 20, 2017

Hmmm, not many, but I don't know for sure. I changed my script to work around the missing v3 features and I didn't track how many times this occurred. I can probably make a quick script to check when my compilation job is complete later this morning.

@hsosik
Copy link
Owner Author

hsosik commented Nov 21, 2017

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants