LDoubleZhi

The problem in train on my own voc data

7

Hi, I use my data like voc dataset in this project.and I have met some problem. First, I replace the offical voc data(20 classes) by my own data(5 classes). and...

论文用三个可学参数作为融合权重，但这3个参数如何学到？

2

论文中没有看到具体如何监督学习这3个参数，希望作者解惑

A problem about anchor setting

3

yolov3用自己的数据集训练时，设置anchor时候需要尺度不变放缩到一个input_size然后聚类得到anchor吗？比如我都resize到640*640，然后kmeans聚类得anchor. 但实际上我debug时候发现源码训练里加入了random_input_size，那这样的话，上面设置的anchor_size就会出现不匹配的情况了(训练时候输入尺度是随机的)． 　　所以应该用多尺度数据聚类anchor吗？

The time consuming of preprocessing?

2

`lb = vn_data.transform.Letterbox(hyper_params.network_size) it = tf.ToTensor() img_tf = vn_data.transform.Compose([lb, it]) s1 = time.time() data = img_tf(img).cuda() e1 = time.time() data = torch.unsqueeze(data, 0) print('preprocessing per img consume: {}'.format(e1 - s1))`...

I'm in china, so you know....

I'm in china, so you know....

## Question