Lab3

HOG 是一种手动算梯度的传统特征提取方法。

HOG

Questions:

• How to understand the log from the logistic regression model?
  • 可以看出来龟速. 另外 eval 准确率只有 0.2038
• How to understand the t-SNE visulizations?
  • 混一块儿没法分开

CNN

Questions:

• How is the classification performance compared to our previous case from Section 3? And why do you think the change happens?
  • 0.3940，这个特征提取更准确
• What can we know about the t-SNE in this case?
  • 看起来好多了

还没写完，先跳过