search

jkjung-avt / yolov4_crowdhuman

A tutorial on training a DarkNet YOLOv4 model for the CrowdHuman dataset
MIT License
139 stars 49 forks source link

Data Preparation #31

Closed mnurilmi closed 2 years ago

mnurilmi commented 2 years ago

hi, thanks for your repo. This is very helpful. However, I still don't understand about the purpose of Kmeans on your data preparation. Does it affect the quality of the dataset?

mnurilmi commented 2 years ago

anw, I've tried using this dataset with yolov7, but the result of objectness loss is sucks. Apart from that, the results are quite satisfactory. My assumption is in the dataset. With the same configuration, I used a bit of the person image dataset and objectness loss just fine. Hopefully u read and respond to this question. thank you

jkjung-avt commented 2 years ago

hi, thanks for your repo. This is very helpful. However, I still don't understand about the purpose of Kmeans on your data preparation. Does it affect the quality of the dataset?

The K-means is just an experiment by myself. I would suggest you to follow the instruction of the original yolov4 repo (AlexeyAB/darknet):

_To calculate anchors: ./darknet detector calc_anchors data/obj.data -num_of_clusters 9 -width 416 -height 416_

mnurilmi commented 2 years ago

ok thanks for the response...for now, I'm using your code and adapting some instructions. So I totally depend on the yolov7 algorithm to do the autoanchor... but overall, my code is based on your repo pipeline