D-Nilsson / GRFP

Code for the CVPR 2018 paper "Semantic Video Segmentation by Gated Recurrent Flow Propagation"
MIT License
56 stars 8 forks source link

Re-training/Fine-Tuning results in bad output #13

Open sbhatia4 opened 3 years ago

sbhatia4 commented 3 years ago

Hi,

ReadMe says about Training as following, but in reality, I don't see the same output after re-training either with Training or Val dataset of cityscapes.

To give you an idea: IoU Class number using pre-trained network - 73.5, whereas, if I train with Training Cityscapes dataset - IoU class number drops to 38.2 while evaluating the Val dataset of Cityscapes.

Any help would be appreciated. Thanks a lot!!

Training Train and evaluate a model with the following commands:

python train.py --static lrr --flow flownet2 python evaluate.py --static lrr --flow flownet2 --ckpt lrr_flownet2_it10000 This should match the performance of the pre-trained LRR model above. See the ./checkpoints directory where parameters are saved during the training procedure. Only LRR is supported at the moment.