ai-cfia / nachet-frontend

Frontend application for seed classification of images acquired from digital microscopes
MIT License
2 stars 2 forks source link

Issue with display?: Numbers in unusual places #102

Open TaranMeyer opened 2 months ago

TaranMeyer commented 2 months ago

Description When I took a picture with multiple seeds, results were mostly correct but the display of the labeling numbers 2 and 4 was odd (they are positioned between the two boxes, not each on top of their respective box)

Expected behavior The model will place a box around each seed, with a labeling number above it

Actual Behavior The labels for boxes 2 and 4 are located between the two boxes, and overlap in some views. Also something is odd with box 8, two seeds close together are bounded in the box and the label looks as if there are two numbers superimposed there, as well.

Supporting Images To better understand the issue, please add screenshots: Results: image

rngadam commented 2 months ago

@MaxenceGui please provide explanations when moving issues around layers.

MaxenceGui commented 2 months ago

This issue is related to the display of the index, which is a frontend issue instead of a model issue. It's the frontend's responsibility to display the results return by the model.

ChromaticPanic commented 1 month ago

Looks like this is Fixed #124

Image

MaxenceGui commented 1 month ago

Kind of hard to read this one: image

What do you think?

ChromaticPanic commented 1 month ago

I see, one easy solution would be to draw a white square background behind any text

then have toggle for labels on or off to see the seeds

MaxenceGui commented 1 month ago

Like that!

TaranMeyer commented 1 month ago

Some kind of toggle for labels would be nice in a lot of situations anyway---we have talked before about having it highlight selected labels and things like that. Often you want to see quickly which seeds a particular label was applied to.

ChromaticPanic commented 1 month ago

Some kind of toggle for labels would be nice in a lot of situations anyway---we have talked before about having it highlight selected labels and things like that. Often you want to see quickly which seeds a particular label was applied to.

Unrelated but just a thought Having a cutout view is possible if a segmentation step is added image

If used as a preprocessor in the pipeline then background color effects on predictions could be eliminated

Having cutouts also brings endless presentation possibilities

MaxenceGui commented 1 month ago

I think you just open a pandora box 😆 @amirardalan9473

TaranMeyer commented 1 month ago

The videometer does a kind of cut-out view thing. It is attractive but does have some problems with cutting off bits, or if seeds touch/overlap (since they're not manually placed). Definitely something we could keep in mind for the future. 😅

ChromaticPanic commented 3 weeks ago

image

136

image