Meet Shah comments

Results 19 comments of


                                            Meet Shah

Overflow encountered in exp after 10 iters, and Segmentation fault(core dumped) after 40 iters.

Check my comment [here](https://github.com/rbgirshick/py-faster-rcnn/issues/65#issuecomment-366613724)

The metrics of computing confusion matrix might be more efficient

Thanks! Can you send a PR?

Any method test validate data on pspnet using pretrained caffe model?

Not at the moment. The PSPNet weight loading is buggy, I'll add these methods once that is fixed.

Size inconsistency in U-Net implementation.

I'm aware of this issue, U-net implementation doesn't support all resolutions. I need to fix this.

Dataloaders for Berkeley's self driving car dataset BDD100k?

That would be great! Looking forward to your PR.

compile error of mxnet/operators

Any solutions to this? I'm using CUDA 8.0 and cuDNN 6.0.21, I still get the error.

compile error of mxnet/operators

I've solved this by using cuDNN 5 with CUDA 8.0.

In training, why the loss is decreasing, while the val loss is increasing?

I'm observing the same phenomena. Is there some fix to this? @kitterive - What initial weights are you using? Also how are you normalizing the coordinates?

In training, why the loss is decreasing, while the val loss is increasing?

@oarriaga - In that case, which model weights did you use to finetune it VGG16 ( with top removed) or the caffe converted SSD weights ?

In training, why the loss is decreasing, while the val loss is increasing?

@oarriaga @rykov8 - Has anyone successfully tried to train the SSD from scratch ( i.e using only VGG16 weights) using this code ? If not then perhaps, it would be...