Neuronal Network with Keras

I have been looking for neuronal networks with keras for objects detection. Currently the ssd networks are having great success, so I’ve tried these networks. For this I have based on: https://github.com/pierluigiferrari/ssd_keras . I did some tests.

1º Test

I tried the ssd300 with 1000 steps per epochs, 70 epochs and adam. I used the parameters, which are in the ssd300 example and I get the next result:

2º Test

I tried the ssd300 with 1000 steps per epochs, 120 epochs and SGD. I used the parameters, which are in the ssd300 example and I get the next result:

3º Test

I tried the ssd7 with 1000 steps per epochs, 20 epochs and Adam. I used the parameters, which are in the ssd7 example and I get the next result:

4º Test

I tried the ssd300 with 1000 steps per epochs, 70 epochs and Adam. In this case the steps parameter is None, the offsets parameter is None and I used the coco scales.

5º Test

I tried the ssd300 with 1000 steps per epochs, 120 epochs and Adam. In this case the steps parameter is None, the offsets parameter is None and I used the coco scales.