tub-rip / event_based_optical_flow

The official implementation of "Secrets of Event-based Optical Flow" (ECCV2022 Oral and IEEE T-PAMI 2024)
GNU General Public License v3.0
144 stars 13 forks source link

flow output #20

Closed valerdi closed 1 year ago

valerdi commented 1 year ago

Hi there, I've runned

python3 main.py --config_file ./configs/mvsec_indoor_no_timeaware.yaml

without changing the original config you provide.

Seems like it is computing optical flow between ind1 and ind2 but is not iterating through the full dataset. Is there a parameter in yaml config to do it? Also, is there a config to output the actual values of optical flow in a file?

One last question, in the paper the multi-scale approach use the information from previous batch to initialize the values of the different scales in the next batch, is this implemented?

Thanks in advance

shiba24 commented 1 year ago

Hi @valerdi ,

  1. Try running with --eval option.
  2. Not in config file. But you can tweak in the visualization part of the solver, settig save_flow=True. Or, you can also save the flow in the main.py after the estimation is done.
  3. Yes. This is the iteration: https://github.com/tub-rip/event_based_optical_flow/blob/main/main.py#L52
valerdi commented 1 year ago

Thanks for the response. I'm playing around with the code, give me some days before closing this to see if I understand well the function you linked.

JulianKrismer commented 1 year ago

Hey, I was trying the --eval option, but got the following error:

2023-10-02 15:05:41,007 - src.feature_calculator - WARNING - Feature calculation is disabled in this source code. 2023-10-02 15:05:41,007 - __main__ - INFO - Sequential optimization Traceback (most recent call last): File "main.py", line 157, in <module> assert loader.gt_flow_available # evaluate with GT flow AssertionError

Is it only possible to run it with ground truth, or is there a way around it?

Kind regards and thank you for your time!

shiba24 commented 1 year ago

@JulianKrismer Hi. You can run without GT flow (with modifying the main.py.) Without GT flow and evaluation, actually one does not need to follow the dt=1,4 conventions in MVSEC, and you can run the flow estimation for arbitrary timestamps. What is your usecase? Do you have specific timestamp list that you want to estimate?

valerdi commented 1 year ago

Hi @shiba24, regarding my questions, everything is running well, thanks a lot!

shiba24 commented 1 year ago

How about @JulianKrismer ?

JulianKrismer commented 1 year ago

I wasn't able to test it yet, unfortunately. I'm working on a master thesis, and am trying to see if optical flow could help me locate the logo of a spinning ball.

shiba24 commented 1 year ago

Ok, let me know if I could help you. Feel free to open another issue if necessary. Closing it as the original question is resolved.