Hello,
According to the method of docker submission in Wiki, the organizers run a docker image as followings.
sudo docker run -it --gpus all --ipc=host -v :/input_video \
-v :/input_kinematic -v :/output \
python /root/MISAW.py /input_video /input_kinematic /output
In this case, the input arguments are "folders" (not "files"). That means the python script (MISAW.py) should analyze ALL files in the "input_video" folder. Is it right ?
Created by SKJP Hello,
The "input_video" and "input_kinematic" folders only contain the testing part. The docker only performe the testing part. So it has to load your model with the weights obtained during the training part. Hi,
Dose the "input_video" folder contains all the video files? Including training and testing videos?
Thanks
LONG Yonghao Yes it is possible to submit one docker per task. In this case please specify the task in the docker image name.
Hi,
Can we submit a separate docker image per task? In our case, we have significantly different models per task and there is no natural means of combining them into the same docker image.
Thank you,
Laura Bravo Hi,
Yes, the input arguments are 3 folders: one for the videos, one for the kinematics, and one for the output.
The python script must analyze **all** files in input_video and kinematic_video folders. The correspondence between video and kinematic can be easily done by their name. For example, for the video "4_1.mp4", the corresponding name for kinematic is 4_1_kinematic.txt.