testing more:
07/4/3
124 frames, including putting 3 plates on, 1 bottle of oil, 1 can of salt
If the weight sensors detect weight changes first, and then check shot detection, it’s possible to improve the accuracy for cooking activity recognition (nutrition detection). It has good ability to filter the cutting action (from frame 67-88, my hand is knocking at the can of salts). However, there are still some issues and exception to be considered.
- check the accuracy
- check exceptions (weight changes but no shot)
- threshold?
- time range?
- check related work in previous post (their methods)
Subscribe to:
Post Comments (Atom)
2 comments:
so you had found some potential breakthrough?
facing the (multimodal) fusion problem? Usually, you can try and start with the "AVERAGE" fusion first. However, the scores of each detector should be normalized (to [0,1]) first.
right... that's fusion problem... I'll try the ave fusion after coming back from CHI07. Thank you so much!!!
Post a Comment