[Bug] Offline Edge Inference flags non-fish objects as "Fresh"

[jpdevhub]

Description

Currently, when a user uses the app in offline mode, the ONNX model is used for edge inference. Unlike the cloud API which validates whether the image contains a fish, the ONNX model forces every image into 3 categories (Fresh, Moderate, Spoiled). Because of this, taking a picture of a human face or a non-fish object often results in a "Fresh" classification.

Tasks

• In src/fusionInference.js, locate NOT_A_FISH_THRESHOLD.
• Increase this threshold (e.g., from 0.36 to 0.55 or higher) to make the model stricter, so it rejects low-confidence non-fish images.
• Alternatively, implement a lightweight object detection step to verify a fish is in the frame before running inference.

[sks-007]

Hi @jpdevhub
I'm an SSOC contributor and would love to work on this issue. Could you please assign it to me?

[Excalibur677]

Hey @jpdevhub, just following up! I'll bump NOT_A_FISH_THRESHOLD to 0.55 in src/fusionInference.js to stop edge inference from misclassifying non-fish objects, and make sure the UI handles the rejected state cleanly. Please assign this to me!

Quick Plan

• Modify: Open src/fusionInference.js and change NOT_A_FISH_THRESHOLD from 0.36 to 0.55.

• Verify: Ensure that if the confidence score is below 0.55, the code triggers a "No fish detected" message instead of defaulting to a category.

• Test: Run npm run dev offline and upload a non-fish image to confirm it gets rejected.