Repository files navigation

author : oukohou
time : 2019-09-26 16:44:48
email :oukohou@outlook.com

the official keras version is here:SSR-Net

• This SSR-Net model can't fit big learning rate, learning rate should be smaller than 0.002.otherwise the model will very likely always output 0, me myself suspects this is because of theutilizing Tanh as activation function.
• And also: Batchsizecould severely affect the results. A set of tested params can be :

  batch_size = 50input_size = 64num_epochs = 90learning_rate = 0.001 # originally 0.001weight_decay = 1e-4 # originally 1e-4augment = Falseoptimizer_ft = optim.Adam(params_to_update, lr=learning_rate, weight_decay=weight_decay)criterion = nn.L1Loss()lr_scheduler = optim.lr_scheduler.StepLR(optimizer_ft, step_size=30, gamma=0.1)

• The dataset preprocess is quite easy. For MegaAsian datasets, you can use the./datasets/read_megaasina_data.py directly;for other datasets, just generate a pandas csv file in format like:

  filename,age1.jpg,23...

is OK. But also, remember to change the./datasets/read_imdb_data.py accordingly.

thanks toDefTruth 's implementation here:How to convert SSRNet to ONNX and implements with onnxruntime c++.

my reading understanding of SSRNet can be found:

• on myblog site here:论文阅读-年龄估计_SSRNet
• or onzhihu here:论文阅读-年龄估计_SSRNet.

which was written in Chinese.