petertor / sparse_convolution Goto Github PK

Dear author,

thank you for the tensorflow version of our paper Sparsity Invariant CNNs.
We would actually like to link your code on the KITTI homepage - would this be fine with you?

Thanks!

Nick

Good Afternoon,

I'm would be very interested in trying this code on the KITTI dataset that comes with the paper on which this code is based, but unfortunetely I'm not quite comfortable with the way to use it.

Would it be possible for you to provide an example code? Or at least guidelines to fullfil this goal?

I thank you in advance for your concern,

Pierre

Hello, I'm reproducing the results in your paper now. I have some questions about the loss function of the network. How can I design the loss function? You have mentioned 'our approach using 5% sparsity during training and testing.', this means the inputs images had 5% sparsity, then the outputs of the network will be more density. What should I compare with the output to get the monitoring signal?
If comparing the output with the input, the predictions will be sparse again?
So I'm so confused. I really hope that you can answer my question. Looking forward to your reply ! Thank you!

Hi,

Why is the max_pooling size is fixed to 3 instead of being kernel_size?

Thanks!

Greetings,

as far as I understood it, calling a convolution with a 1-filled matrix and then inverting the result on the mask should result in a decrease of depth values at attention boundaries. Example:

5 visible pixels on an 3x3 kernel results in the value 5 in the intermediate mask(IMask), resulting in 0.2 after the reciprocal operation.

Anyways, results look great!

Hello. For the kitti depth dataset, can I get the intrinsic parameters of the camera which record the images ? thanks!

Hello,
I has completed the training and testing for depth completion with your sparse convolution code, but the result show that the error of D1-all increase rapidly when we decrease the density of input depth. I want to known your experimental result and if you can offer more guidance, please contact we with email. My email is [email protected].