search

SoonminHwang / rgbt-ped-detection

KAIST Multispectral Pedestrian Detection Benchmark [CVPR '15]
http://multispectral.kaist.ac.kr
Other
312 stars 63 forks source link

Where can I find the reasonable subset? #2

Open Sherry40931 opened 6 years ago

Sherry40931 commented 6 years ago

Hi, I just downloaded the dataset but I could not find the reasonable subset mentioned in your paper. Is it one of the image lists in the imageSets subdirectory? Also, could you tell me the relation of the benchmark mentioned in your paper with the lists in the imageSets directory? Thank you very much!

Sherry40931 commented 6 years ago

I think I have got the point. For those who are encountering the same problem, the reasonable subset consists of not/partially occluded pedestrians which are larger than 55 pixels.

As far as I know, no-occlusion is tagged with 0, partially-occlusion is tagged with 1 and heavy-occlusion is tagged with 2. Please correct me if I am wrong, Thank you very much!

anushl9o5 commented 5 years ago

Hi, I still don't get how you decide to ignore a bounding box from the subset? If you have understood please explain. Thanks

MuhammadAsadJaved commented 4 years ago

I think I have got the point. For those who are encountering the same problem, the reasonable subset consists of not/partially occluded pedestrians which are larger than 55 pixels.

As far as I know, no-occlusion is tagged with 0, partially-occlusion is tagged with 1 and heavy-occlusion is tagged with 2. Please correct me if I am wrong, Thank you very much!

Thank you for this explanation. It's very helpful. I have a few more questions about the dataset.

1 - There are 11 sets in total with several subsets.
Some .txt annotation contain only " % bbGt version=3" this tag and no bounding box values. What does this mean? Should we remove these annotations?

2- Can you explain the values in the .txt file? for example annotation set00/V000/I02165 contains these values. The first one is class, next 4 are bounding box coordinates ,what about other values? Do we need these in the training ?

% bbGt version=3 person 427 243 27 66 0 0 0 0 0 0 0

3- How we divide these annotation to 9 categories? Reasonable all, Reasonable day, Reasonable night, Near scale, Far scare, Medium scare, No occlusion, Partial occlusion, and Heavy occlusion? Anyone write any scripts to categories images and .txt files? Please also see attached images for Question 1 and 2.

Thank you very much.

kaist1 kaist2