Fig.1.1. Regression Effect on Test Data (image is input, label is regression result)
From Fig. 1.1 and Fig. 1.2, we can see that the labels of the regression are all right, the generation is ok although it is a little blurry. Through test, we can conclude that we can use one neural network to recognize multi-attribute so long as the attributes can be represented by functions.
Generation Test
Now let’s modify the label to some degree, then see the generation. First, I change the color to [0.5,0.5,0], [0.5,0,0.5], [0,0.5,0.5], [0.3,0.3,0.3], [0.5,0.2,0.3], [0.2,0.3,0.5], [0.3,0.5,0.2].The result is shown in Fig. 2. Second, I change the size to[0.5,0.5], [0.3,0.7], the result is shown in Fig. 3.. Last, I change the location to[0,1.05], [1.05,0], [0.5,0.5], [0.55,0.45], the result is shown in Fig. 4. Why the sum is equal to1? Because the label is one hot encoding and the attribute is exclusive, we can’t make one thing big and small simultaneously. The connection can’t make the output to 1 simultaneously.