I have some newbie questions. I used the darknet19 model to load the corresponding pre-trained weights from the Darknet website (ImageNet 1000 classes). I see the last softmax layer of this model is of shape (None, 8, 8, 1000).
How do I use this for simple classification? The prediction of one of the sample images (eagle.jpg) is of shape - (1, 8, 8, 1000). How do I interpret this? Do I look for the max value of the 1000 elements across the 8x8 3D matrix?
Could this be flattened into 1x1000 array? Would that require re-training?
I want to retrain for a new dataset of 200 classes. Change I change the last softmax layer to shape 200 and train by freezing the other layers?
I have some newbie questions. I used the darknet19 model to load the corresponding pre-trained weights from the Darknet website (ImageNet 1000 classes). I see the last softmax layer of this model is of shape (None, 8, 8, 1000).
How do I use this for simple classification? The prediction of one of the sample images (eagle.jpg) is of shape - (1, 8, 8, 1000). How do I interpret this? Do I look for the max value of the 1000 elements across the 8x8 3D matrix?
Could this be flattened into 1x1000 array? Would that require re-training?
I want to retrain for a new dataset of 200 classes. Change I change the last softmax layer to shape 200 and train by freezing the other layers?