Hi,
Thanks for your code, and after I ran the slow network to predict the trainingH datasets, I got much worser results than the results displayed on the Middlebury website, I wonder where I did something wrong, can you help me out ?
The command is as follows:
./main.lua mb slow -a predict -net_fname net/net_mb_slow_-a_train_all.t7 -left im0.png -right im1.png -disp_max 380
And one of the results can be compared as follows: The error rate(nocc) for Vintage by running the code is 31% while the result of Vintage image on Middlebury (MC-CNN-acrt ) is 24.8%.
Do I miss something important? Thanks a lot !
Hi,
Thanks for your code, and after I ran the slow network to predict the trainingH datasets, I got much worser results than the results displayed on the Middlebury website, I wonder where I did something wrong, can you help me out ?
The command is as follows:
./main.lua mb slow -a predict -net_fname net/net_mb_slow_-a_train_all.t7 -left im0.png -right im1.png -disp_max 380
And one of the results can be compared as follows: The error rate(nocc) for Vintage by running the code is 31% while the result of Vintage image on Middlebury (MC-CNN-acrt ) is 24.8%.
Do I miss something important? Thanks a lot !