How well can this perform on outdoor scenes - roads for example?

[MeghaMaheshwari]

No description provided.

[kashgupta]

Should work well for outdoor scenes as long as you're looking to detect depth of an object in the scene. The algorithm only draws bounding boxes around objects and assesses depth within that bounding box. It would be hard to classify the depth of the road but easy to classify the depth of a car.

[MeghaMaheshwari]

I see that you used numpy array files for training. What are these matlab contents from which you generated the .npy files? I would like to train the model on custom data and see the output. I am having issues getting the input ready. I have image files and the corresponding depth files as png. How can I get the training set ready to feed into the network.

Also how many images did you use for training?

[kashgupta]

Those npy files are actually just packaged numpy filed that we got from NYU's public dataset to train the model on. In order to train on custom data you would want to use some package to convert png images into numpy arrays (3-dimensional), and then you would feed those input arrays. The image file will be RxCx3 and the depth file will be RxCx1. In all honesty our code would not work for another application out of the box and it is better just to read our pdf and implement a similar method.

…

On Mon, Dec 3, 2018 at 4:09 PM Megha Maheshwari ***@***.***> wrote: I see that you used numpy array files for training. What are these matlab contents from which you generated the .npy files? I would like to train the model on custom data and see the output. I am having issues getting the input ready. I have image files and the corresponding depth files as png. How can I get the training set ready to feed into the network — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ARH6cLnb8cog6jESKjfv7Qz1pIGcKMSRks5u1b0egaJpZM4YrqlS> .

[MeghaMaheshwari]

Do you have a link to the paper on which this idea is based?

[kashgupta]

It should be in the references section of this paper: https://kashgupta.com/depth_detection.pdf

…

On Mon, Dec 17, 2018 at 2:59 PM Megha Maheshwari ***@***.***> wrote: Do you have a link to the paper on which this idea is based? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ARH6cHnb_IsuQtn9deCwnKVekzAwMa8Kks5u6CHjgaJpZM4YrqlS> .