Help us improve the model

1import lynx
2 
3# From a file path
4lynx.submit_feedback("frame.jpg", expected="should detect 2 dogs")
5 
6# From a Results object — source image attached automatically
7results = model("frame.jpg")
8results.submit_feedback(expected="this is a deer, not a horse")

The model surprised you. Maybe it missed a detection that's obviously there. Maybe it labeled a deer as a horse. Maybe it returned three boxes where there's clearly one. Send us the image plus a short note about what you expected — the next training cycle adds it to the curation queue and the model that ships after that learns from it.

Use it for:

• Real failures you encountered in your own data — these are the highest-value training examples because they represent the deployment distribution, not the training set
• Edge cases you can't reproduce locally — submit them while you have the frame; reproducing later is harder than you think
• Cases where autotune helped — pair the two recipes: if autotune found a working config that your default doesn't use, include that in the expected note ("autotune says BGR works; default RGB returns nothing")

Anonymous submission is accepted — no API key required. Passing key= or setting LYNX_API_KEY attributes the feedback so we can follow up with you specifically when the model fixing your case ships.

Numpy and PIL inputs are auto-encoded to PNG before upload, so we always receive a known-format byte stream regardless of what you had locally. Returns {status, feedback_id, body} — keep the feedback_id if you want to reference it in a support thread.