search

BUT-GRAPH-at-FIT / PersonGONE

Implementaion of PersonGONE
2 stars 2 forks source link

detect ROI Issue --- 'NoneType' object has no attribute 'shape' #1

Open txy8496 opened 2 years ago

txy8496 commented 2 years ago

Dear Team,

I am interested in the performance of Track 4 of AI CITY CHALLENGE and want to find more details of the contest.

During re-running your code on data TestB, I have downloaded the pre-trained models and successfully finished the in-painting process. While when I run Detect ROI on TestB, there are following errors:

image

According to the image, it said "it can not open PersonGONE/data/mean_scenes/testB_1.jpg". When I checked the files, I found there did not exist the files named PersonGONE/data/meanscenes/testB*.jpg. That is, there did not exist the images in the folder as shown below. I have tried the whole procedure multiple times and the error still occurs.

However, I ran the model perfectly with no error on TestA because there exists these files like "PersonGONE/data/mean_scenes/testA_1.jpg" in the folder as follows.

image

How can I find those missing files for TestB?

I appreciate if you can help me figure this out. Thank you for your help.

Sincerely, Xinyu

VojtechBartl commented 2 years ago

Dear Xinyu,

there are two sub-processes as a part of Detect ROI process - extracting mean background model and ROI detection. First of these probably failed. The background extraction is very memory consumpting and it sometimes does not finish correctly. You can try to decrease the count of processed frames in file _utils/bckgsubtraction.py on line 19 to something like for _ in tqdm(range(3000)): Decreased count of proceeded frames can lead to slightly worse mean background image, but should be sufficient.