search

torrvision / crfasrnn

This repository contains the source code for the semantic image segmentation method described in the ICCV 2015 paper: Conditional Random Fields as Recurrent Neural Networks. http://crfasrnn.torr.vision/
Other
1.34k stars 462 forks source link

Training on Pascal VOC #153

Open ankitsanand opened 6 years ago

ankitsanand commented 6 years ago

Hi Everyone

I am trying to replicate the training of this model on Pascal VOC augmented dataset which has ~11500 training images and 346 validation images.

I have following queries regarding training:

  1. For me, a whole pass through the dataset of 11500 images takes approximately 15-16 hours where i use a batch size of 1. Am I missing something or is it similarly slow for everyone?
  2. How many complete passes/epochs through complete datasets are required to train model to desired accuracy roughly 70 Miu?
  3. How is the trend - Will initial validation Miu accuracy decrease starting from fcn8 and then decrease. Any trend in this whosoever has trained successfully will really be of help.

Regards Ankit