Post-classification after inference



I designed a fire detection using Deep Learning based classification approach. In my training dataset, I have both fire and fire smokes are supposed to be detected (all under "fire"; mostly real fires are detected. Fire smokes are less accurate).

Now after months, I need to differentiate them in my detection results. It would be difficult to retrain each class separately now. Another option coming into my mind is building a binary classification after the main one, getting the main detections as input and saying which of the two it belongs to. However, I may miss some fire smokes I believe because that's less accurate.

Is there any other approaches? What are pros/cons of various approaches?

Tina J

Posted 2020-02-04T16:40:39.650

Reputation: 889

Try just fine-tuning your model by only retraining the classification head / last layer. – mshlis – 2020-02-04T16:52:21.050

No answers