search

ai-cfia / nachet-frontend

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

User Reported Bug: Seed missed for some reason? #132

Open TaranMeyer opened 1 month ago

TaranMeyer commented 1 month ago

Describe the bug Complete seed in image missed/skipped, no bounding box or ID returned. Oddly the program found and ID'd (correctly) the second, partial seed on the right hand side of the image.

To Reproduce Steps to reproduce the behavior:

  1. Load image shared below (.jpg) into model
  2. Analyse with Swin Transformer
  3. Results returned

Expected behavior The complete seed in the middle of the image would have a red bounding box and label corresponding to its ID. The partial, cutoff seeds might or might not be identified and/or ID'd correctly.

Screenshots

Image

Image

Desktop (please complete the following information):

Additional context The new model is usually pretty good with segmenting out the seeds, I'm not sure why it would miss this one as it seems pretty well defined to me.

MaxenceGui commented 1 month ago

Hi @Taran, I can confirm that the model is returning 2 predictions. However, one of them is below 50% meaning that it won't currently show on the deployed version. I know @ChromaticPanic and @CFIALeronB have work to change that feature.

☝🏻 When they add this change to Nachet, it should show the missing box.

TaranMeyer commented 1 month ago

Awesome, thank you! Good to know why this happens sometimes.