camel007 / caffe-yolo-face-detection Goto Github PK

Hi
I have tried your model to detect faces on the UnConstrained College Students (UCCS) dataset (http://vast.uccs.edu/Opensetface/) , but is performing very poorly.

In total: 11110 faces, 70 detected faces and 3466 false alarms
Accuracy: 0.0063006300630063005

I have tried to change the overlap_thresh = 0.5; to 0.4 in yolo.cpp. but still not much improvement. Is there any other parameters that I can tune to get improved results?

Thanks in advance.

Hei, thanks for your share~ I have a question is that in each cell ,it have (1+4+20)labels, if a cell is responsible for an object, it's label is easy to know, but,once a cell is not a center of an object ,,how to set the value of labels ? need your help~

@camel007 Hi. I want to train YOLO on my own dataset, but I don't know how to use your opensource.

1. Is leaky layer needed to add in the caffe/layers?
2. Should I compile the pycaffe in advance, and then config it in my Anaconda to train model?

Hi，can you show me how to prepare the train data as the format of lmdb
That is, I don't know the format in train.txt

解决了

Hi,
what's your database?Is public or private?If it's public,could your provide the link?Thx.