Running on custom datset

NVlabs / SENSE

SENSE: a Shared Encoder Network for Scene-flow Estimation

Other

78 stars 14 forks source link

Running on custom datset #6

Open aakash26 opened 4 years ago

aakash26 commented 4 years ago

Hi Authors,

Thanks on providing the code. Is it possible to run the code on custom stereo video sequences.

Regards Aakash

playerkk commented 4 years ago

Hi Aakash, Stereo videos are sufficient for stereo disparity and scene flow estimation.

aakash26 commented 4 years ago

HI @playerkk ,

Thanks for the quick reply. Yes I understand that. However just to confirm, the stereo camera calibration is required right?. Also to run for custom video sequence, I can place my stereo pair in /data/image_2 andrun demo.py ?

playerkk commented 4 years ago

It is assumed that two stereo images have been rectified.

You can simply run demo.py with your own images.

aakash26 commented 3 years ago

Hi @playerkk ,

yes while trying to run demo.py with given datset I am getting this error(image attached) I have made no changes in code or given sample dataset

Regards Aakash Rajpal

playerkk commented 3 years ago

Hi @aakash26, can you provide more information of your experiment environment? Especially PyTorch version. Thanks.

aakash26 commented 3 years ago

Hi @playerkk, env

Thanks for the reply. I have attached the screenshot of my python 3.6 environment with torch version 1.3.0.

Regards Aakash Rajpal

aakash26 commented 3 years ago

Hi @playerkk,

I was looking into the error and the code. I found that the forward module defined in file SceneFlow\SENSE\sense\models\models.py expects 4 image arguments with cur_im beign the only postional argument.

forward error

However, when calling the Model in SceneFlow\tools\demo.py we are only passing 3 images to the Network.

holis I am not sure if the error is caused due to this or it's something completely opposite.
Thanks for your time

Regards Aakash Rajpal

playerkk commented 3 years ago

Hi @aakash26, are you running demo.py with your own data or the provided images in this repo? Have you made any modifications to demo.py?Thanks.

aakash26 commented 3 years ago

Hi @playerkk,

Currently I am running on the original images provided with no changes in demo.py. I am running all the original scripts.

Regards Aakash Rajpal

aakash26 commented 3 years ago

Hi @playerkk,

I was able to make it run by giving positional keywords for each image in demo.py (screenshot attached) but the results don't seem that good. However, I wanted to understand the order in which the images are declared in the forward function, like what should be passed for cur_im, nxt_im , right_im , left_im