search

yuanyuanli85 / Stacked_Hourglass_Network_Keras

Keras Implementation for Stacked Hourglass Network
112 stars 44 forks source link

Training on a custom dataset #10

Open pavan4 opened 5 years ago

pavan4 commented 5 years ago

I want to train this network on a custom dataset. I have the gt labels and images.

Under the data_gen directory I looked at MPIIDataGen but this class has a lot of custom code specific to MPII. Is there any other place I need to modify apart from this class?

yuanyuanli85 commented 5 years ago

basically, you need to write your own data generator. load data, preprocessing, data agumentation...etc.

pavan4 commented 5 years ago

Ok. I am down this path. I just have a clarification question

When processing data a scaling factor of 200 is used throughout.

Eg.,

https://github.com/yuanyuanli85/Stacked_Hourglass_Network_Keras/blob/eddf0ae15715a88d7859847cfff5f5092b260ae1/src/data_gen/data_process.py#L10

https://github.com/yuanyuanli85/Stacked_Hourglass_Network_Keras/blob/eddf0ae15715a88d7859847cfff5f5092b260ae1/src/data_gen/data_process.py#L48

Is this specific to the MPII dataset? Apologies for the noob question

yuanyuanli85 commented 5 years ago

yes, that is specific to MPII dataset.

YinZ-510 commented 5 years ago

@pavan4 Hi, I also want to train this network on on my own dataset, my input picuture size is 720x1280， but when I modify the mpii_datagen.py I don't know how to set the scale and objpos. Can you give me some advice?