Continual Learning using Mask RCNN ( TensorFlow + Keras) -- 2
Budget: $30 – $250 USD
My project is to add new custom dataset to the original one Incrementally.
For example, I want to add new class while preserving other classes from previous model. The output will be a new class with the original one.
To be clear, I have two classes named ( permanent teeth, primary teeth, and filling).
Model 1 trained on PERM dataset ( single class)
Model 2 used pre-trained from model 1 to train new class (PRIM dataset). The output should be ( two classes, PERM & PRIM).
Model 3 used pre-trained from model 2 to train new class ( filling). The output after this model should be 3 classes and so on.
I want to tackle the problem of catastrophic forgetting.
Requirements:
1. Using Mask RCNN ( TF & Keras)
2. Applied the continual learning approach similar to the attached (class incremental learning)
3. Evaluate the mAP and accuracy for each class once we have more than one class.
For example, I want to add new class while preserving other classes from previous model. The output will be a new class with the original one.
To be clear, I have two classes named ( permanent teeth, primary teeth, and filling).
Model 1 trained on PERM dataset ( single class)
Model 2 used pre-trained from model 1 to train new class (PRIM dataset). The output should be ( two classes, PERM & PRIM).
Model 3 used pre-trained from model 2 to train new class ( filling). The output after this model should be 3 classes and so on.
I want to tackle the problem of catastrophic forgetting.
Requirements:
1. Using Mask RCNN ( TF & Keras)
2. Applied the continual learning approach similar to the attached (class incremental learning)
3. Evaluate the mAP and accuracy for each class once we have more than one class.