Learning the Differences between Softmax and Sigmoid for Image Classification

Your article gives me so many Eureka!s. How are the satellite images got loaded into your Optimized TensorFlow? I have never used TensorFlow so I don't know the pre-processing step. How do you turn the images into blobs? Do you use Opencv? I still cannot detect snakes that don't curl :P

Nice explanation! Can you recommend me some techniques or some papers mentions about how to build a multi-label dataset? This work requires a lot of efforts.