Open ahkarami opened 6 years ago
one calss =>Not really more accurate but faster. Coco is a good dataset for pedestrian.
I also want to train network for pedestrian detection .What should the classes set? classes =1? or classes =2? 2 means that background and person. Please tell me, thank you! @ahkarami @jjehl
@sunshinezhihuo you don't want to detect background so you should not have background included in the class set.
Just classes=1, and update the filters and files related correctly.
It's good to have negative images in your dataset (e.g. related images with no persons at all, then darknet will learn to filter the background away upon detection).
@TheMikeyR OK,thank you very much.
Hi, i would like to train a model just for pedestrian and animals.
Thank you for your great work!
I trained my yolo v3 using coco person and voc person. The loss is below 1. But the detect result is not that good while the pretrained model from the autuor works really well. I don't know why.
Dear @Paul0629, I think your work is valuable. Have you used the default hyper-parameters of YOLO v3? I think, it is better to change the default hyper-parameters for this task. For example change the anchor boxes. You can refer to training Yolo on the Cityscapes dataset
can anyone share better yolo .cfg model filw and corresponding .weights file for better pedestrian detection? Thanks in advance!
What should I do for pedestrian detection alone
Hi, I have a general question. In the VOC and COCO datasets there are some classes (e.g., 20 or 80 classes (person, car, horse, etc.) for object detection). If one train the network just for one class (e.g., just the person class) will the accuracy (mAP) of person detection increase? Another question, does anyone train the YOLO v3 on pedestrian detection data sets (e.g., Caltech)?