相比DeconvNet写得比较简洁易懂。但是原来代码里面给的数据似乎没法跑。所幸同作者还有一些代码用到了TCNN，比如action recognition，可以一起下载下来参考。这个代码主要特色就是tiled结构，可以用来参考，然后把里面ICA的优化函数换成RICA，Sparse Coding等等。总之，在这个代码里也了解了不少(比如line search等)。-DeconvNet well written and easy to understand compared to relatively simple. But the original code to the data inside can not seem to run. Fortunately, there are some code with the author uses TCNN, such action recognition, can be downloaded along down the reference. This code is the main feature tiled structure that can be used for reference, and then put inside the ICA optimization functions into RICA, Sparse Coding and so on. In short, this code can also learn a lot (such as line search, etc.).