Monday, August 29th, 2022 - Saturday, September 3rd, 2022

Two days after I started to do the experiment, I suddenly found that the code repository (BDCNN) I was working in is for the paper [1] which is about micro-expression recognition, not micro-expression spotting.

Then I switched to SoftNet-SpotME which is for paper [2]. The first step is face cropping which cost four days to finish in the SAMM dataset and two days in the CAS(ME)^2 dataset, which really shocked me a lot. The algorithm is not sophisticated. The Python process cost only about 5% of the CPU’s computing ability. It was just because the dataset has a large number of images.

During my experiments, the C disk only left about one G of storage. Holy shit!

After months of researching, I finally found the code that could run successfully.

[1]B. Chen, K.-H. Liu, Y. Xu, Q.-Q. Wu, and J.-F. Yao, “Block Division Convolutional Network with Implicit Deep Features Augmentation for Micro-Expression Recognition,” IEEE Transactions on Multimedia, pp. 1–1, 2022, doi: 10.1109/TMM.2022.3141616.

[2]G.-B. Liong, J. See, and L.-K. Wong, “Shallow Optical Flow Three-Stream CNN for Macro- and Micro-Expression Spotting from Long Videos.” arXiv, 2021. doi: 10.48550/ARXIV.2106.06489.