I'm a newbie in neural networks. I'm trying to fit my neural network that has 3 different outputs:
- semantic segmentation,
- box mask and
- box coordinates.
When my model is training, the loss of semantic segmentation and box coordinates are decreasing, but the loss of the box mask is increasing too much.
My neural network is a CNN and it's based on Chargrid from here. The architecture is this:
For semantic segmentation outputs, it's expected to have 15 classes, for box mask it's expected to have 2 classes (foreground vs background) and for box coordinates it's expected to have 5 classes (1 for each corner of bounding box + 1 for None).
Here's the loss/accuracy for each of the three outputs at the end of the first step.
- Semantic segmentation (Output 1) - 0.0181/0.958
- Box mask (Output 2) - 13.88/0.946
- Box coordinates (Output 3) - 0.2174/0.0000000867
Here's the loss/accuracy at the last step.
- Semantic segmentation (Output 1) - 0.0157/0.963
- Box mask (Output 2) - 73.02/0.935
- Box coordinates (Output 3) - 0.06/0.82
Is that normal? How can I interpret these results?
I will leave below, the output of the model fit.