Embedded Files
Towards Real-Time Generation of Delay-Compensated
Neeloy Chakraborty*, Yixiao Fang*, Andre Schreiber, Tianchen Ji, Zhe Huang, Aganze Mihigo,
Cassidy Wall, Abdulrahman Almana, and Katherine Driggs-Campbell
University of Illinois Urbana-Champaign
Accepted in ICRA 2025
presentation_recorded.mp4Supplementary Video
Supplementary Video
We highlight some results from depth prediction, end-to-end generations, and real-time evaluations on rosbags.
Below, we showcase additional videos of our method
operating in complex under-canopy environments.
We recommend switching from "Auto" to "1080p" under the settings
of each video to view in high-definition in full-screen mode.
Ours Running Real-Time on Early Growth Rosbag
3 FPS - 250 ms Delay
1.mp43 FPS - 250 ms Delay
2.mp43 FPS - 250 ms Delay
3.mp43 FPS - 500 ms Delay
1.mp43 FPS - 500 ms Delay
3.mp4Ours Running Real-Time on Middle/Late Growth Rosbag
6 FPS - 250 ms Delay
1.mp46 FPS - 250 ms Delay
3.mp46 FPS - 500 ms Delay
1.mp46 FPS - 500 ms Delay
2.mp43 FPS - 250 ms Delay
1.mp43 FPS - 250 ms Delay
2.mp43 FPS - 500 ms Delay
1.mp43 FPS - 500 ms Delay
2.mp4Ours Running Real-Time on Late Growth Rosbag
6 FPS - 250 ms Delay
1.mp46 FPS - 250 ms Delay
2.mp46 FPS - 250 ms Delay
3.mp46 FPS - 250 ms Delay
4.mp46 FPS - 500 ms Delay
1.mp46 FPS - 500 ms Delay
2.mp46 FPS - 500 ms Delay
3.mp43 FPS - 250 ms Delay
1.mp43 FPS - 250 ms Delay
2.mp43 FPS - 500 ms Delay
1.mp43 FPS - 500 ms Delay
2.mp4Failure Cases
Poor Depth and Pose Prediction
2.mp4Occlusion Leading to Poor Pose Predictions
4.mp4Random Occlusion in Generation
2.mp4Page updated
Report abuse