search

CharlesShang / TFFRCNN

FastER RCNN built on tensorflow
MIT License
874 stars 420 forks source link

regarding using the code for different data set #107

Open wenouyang opened 6 years ago

wenouyang commented 6 years ago

Hi Charles,

Thanks for sharing the code. If I would like to leverage the code to train on the new data set, what are the modifications that I have to made? Can I still use the pre-trained weight? Thanks.

ChrisDal commented 6 years ago

You could initialize your neural network from the pre-trained weights and then train the network on your new dataset (also called fine-tuning). For the modification, you have to take a look at the way to load the dataset : TFFRCNN/lib/datasets/, you could begin from thepascal_voc.py and factory.py files to build your new dataset load file.

Chris