Prepare data for quantitative analysis

[nlitz88]

• [x] Need to create a rosbag recording the entire pipeline running. This will capture both the intermediate BEVFusion output AND the BEV map with the predicted workzone masks. DONE: Here: https://drive.google.com/file/d/1J2QiT3MSH1UXkZMs6GTJpIXaqSVx7RnP/view?usp=drive_link (shouldn't need to use this, though).
• [x] Will find some kind of utility to export / extract the images from the ros bags for @brendaju to use.
• [x] Upload the images to Google Drive. DONE: Here: https://drive.google.com/file/d/1w5oQI9uWOR0VSK9zqobMvtuqYPXzDHIq/view?usp=drive_link

[nlitz88]

@brendaju Okay, the images are linked above in that third bullet point. Let me know if you are able to access them okay.

[nlitz88]

WAIT--There is one potential problem with these: I don't think we can trust the numbers/indexes of those images. The export tool I used (https://github.com/MapIV/ros2_bag_to_image) had an issue where if your image messages weren't timestamped, then it would bug out--so I had to increment things manually. However, this could lead to some inconsistencies.

To resolve this, the right answer is to properly timestamp our image messages. Not rocket science, but would just probably be a tiny bit of leg work from @CMUBOB97 and myself.

@CMUBOB97 I can look myself, but before I dive in: do you know where the BEVFusion node creates the image_msg published to result_image? I.e., where we might be able to set the timestamp?

Basically, we just need to set the timestamp in the BEVFusion node. THEN, we have to update Chi's node to *set the timestamp of the boundary_image we create to be == the timestamp found in the received result_image's header

[nlitz88]

@CMUBOB97 Update: I just found where they create the image and it looks like they didn't set the timestamp. I just added a fix here in the ros2-galactic branch I made yesterday https://github.com/nlitz88/BEVFusion-ROS-TensorRT/commit/238103446b2834e32b3df733e1d80ce9725586f8, so hopefully we are good/set on the BEVFusion node front.

Now all we need is a quick one-line update to the boundary generation node.

[nlitz88]

@CMUBOB97 Another update: I just updated your node so that the published image's timestamp matches the timestamp of the incoming message. https://github.com/CMUBOB97/ROS2-Boundary-Segmentation/commit/18f3c47e3b27c20b8e80b0d3c4cfe2f2ab5988dc

So in theory, now we should be okay. I.e., now we can know which result_image from BEVFusion truly corresponds to each boundary_image from the boundary segmentation node.

I will re-record those ROS bags and upload them again. Sorry for the delay.

[nlitz88]

Okay, re-recorded, exported, and updated the google drive links above. For now, just have images from Scene 61, but I can make some additional scenes as well.

[CMUBOB97]

Yay!

[CMUBOB97]

@nlitz88 I just pushed a change on boundary segmentation package so that it overlays the ego frame on top of the segmented workzone (boundary_image). Also let me know if you need help in generating the binary prediction results (or maybe you have already acquired it).

[nlitz88]

@CMUBOB97

Okay, I see the ego-vehicle overlaid. Very cool!