XiandaGuo / OpenStereo

OpenStereo: A Comprehensive Benchmark for Stereo Matching and Strong Baseline
423 stars 49 forks source link

question about having large weights after training #147

Open 13185742215 opened 3 weeks ago

13185742215 commented 3 weeks ago

I changed the MAX_DISP in the yaml file from 192 to 256 before training, and the weights after training is 475 MB. We tried training with a maximum disparity of 192, with a weights of 294 megabytes. The pre-trained model you provide is obviously much smaller. Is there a way to compress the model?

By the way, what is the method of continuing training after the training is interrupted?