Recurrent-Scene-Parsing-with-Perspective-Understanding-in-the-loop
Recurrent-Scene-Parsing-with-Perspective-Understanding-in-the-loop copied to clipboard
aimerykong
When I test the demo_NYUv2
Hi, thanks for your share!
I am testing your code on other images of the NYUv2 dataset. ^_^
There is a problem looks like this:
It seems that I don't have the 1.mat. I load your 5017.mat and I can see there are 'segGT', 'img', 'depth', 'depthBin'， the previous three can be loaded from the NYUv2 dataset, but I must set the threshold of depth to generate depthBin, and I want to know how to set the threshold .
Thanks in advance.
Hi, thanks for your interest. I only upload a few examples for demonstration. If you want to test other images, you need to prepare the data in the same format as those .mat files. Otherwise, you can play around the code to let the model receive your image directly.
Setting the threshold is data dependent. In our paper, we note that by studying the distribution of the depth values, we choose the thresholds to quantize into five bins, in each of which there are almost equal number of depth values. This is not the optimal way but it works. If you want to have a clever way to gate multi-scale information without quantizing the depth, you can refer to our another work ( https://github.com/aimerykong/Pixel-Attentional-Gating/tree/master/demo1_Panoramic_Surface_Normal), that trains with an attentional mechanism to select the scales (without depth anymore).
On Thu, Nov 15, 2018 at 1:08 AM myhooo [email protected] wrote:
Hi, thanks for your share! I am testing your code on other images of the NYUv2 dataset. ^_^ There is a problem looks like this: [image: image] https://user-images.githubusercontent.com/18443212/48537737-6689cc00-e8ed-11e8-947a-8f5783782414.png It seems that I don't have the 1.mat. I load your 5017.mat and I can see there are 'segGT', 'img', 'depth', 'depthBin'， the previous three can be loaded from the NYUv2 dataset, but I must set the threshold of depth to generate depthBin, and I want to know how to set the threshold . Thanks in advance.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/aimerykong/Recurrent-Scene-Parsing-with-Perspective-Understanding-in-the-loop/issues/11, or mute the thread https://github.com/notifications/unsubscribe-auth/AGKZJFBfcbeO5NJoFzAfYiliFCpNXvUyks5uvS73gaJpZM4YfXb- .
