I have a dataset of 3D-printed line images with different line thicknesses. I am looking to classify these images. The difference between the lines of two neighboring classes is clearly visible to the human eye, as shown in the example image. Since my dataset is small (300 images per class), I need to avoid large conventional networks, which tend to overfit. I tried a fairly simple convolutional network with conv2d/batch normalization/pooling and finally two dense layers. Any one as an idea Continue reading...
