RCF-pytorch Failed to reproduce results

I failed to reproduce the results on the README using your training script. The scores were

ods: 0.7926  (README says I should get 0.808)
ois: 0.8073   (README says I should get 0.811)

The trained weights can be found here https://drive.google.com/file/d/10N9ohhzJvHdbrYfVd_1zzJ-vp46XEiZN/view?usp=sharing. The code I used is here https://github.com/yuyu2172/RCF-pytorch.

If possible, could you upload the weights of the trained model you used to produce the results on README? Also, could you upload VGG16 pretrained weights?

Oct 22 '18 03:10 yuyu2172

your result is common. The pytorch version's convergence speed is faster than the caffe version's. I get the best result in the training stage, not the last stage, however, I forgert the detail now. I am doing my work for ddl now. So, I will consider update the readme in detail after my ddl. Have a good luck~

Oct 25 '18 05:10 meteorshowers

@yuyu2172 where did you got the vgg16convs.mat file？

Nov 06 '18 13:11 22920132203749

Failed to reproduce results....... I get ods: 0.76 on your @yuyu2172 trained model and 0.75 on mine. (BSDS500, test set)

I use Structured Edge Detection Toolbox to evaluate my results, but I'm not sure the parameters of edgesEvalDir.m, could you please tell me yours? @meteorshowers @yuyu2172

Here is my parameters: thrs: 99 maxDist: 0.0075 pDistr: {'type' 'parfor'} thin: 1

Feb 07 '19 09:02 mayorx

Here are the results by using @yuyu2172 's model.

Here are the results of mine.

Feb 08 '19 11:02 mayorx

Here are the results by using @yuyu2172 's model.

.....

Because it need nms first ... before nms: after nms: Here is an example of the evaluation process.

Feb 12 '19 07:02 mayorx

I failed to reproduce the results on the README using your training script. The scores were
ods: 0.7926  (README says I should get 0.808)
ois: 0.8073   (README says I should get 0.811)
The trained weights can be found here https://drive.google.com/file/d/10N9ohhzJvHdbrYfVd_1zzJ-vp46XEiZN/view?usp=sharing. The code I used is here https://github.com/yuyu2172/RCF-pytorch.

If possible, could you upload the weights of the trained model you used to produce the results on README? Also, could you upload VGG16 pretrained weights?

Hi~ Have u done any modification, such as learning rate, the type of optimizer, or data augmentation? Because I got wrong results where the last two ouput is totally black. Thanks so much for ur kind help.