pytorch-retinanet

pytorch-retinanet copied to clipboard

About detecting many extra boxes

2

Hello author, thank you for your hard work! Drawing on your ideas, I am developing my own single-stage object detection network, but there are a lot of extra boxes in...

pengjizhi

There are too many prediction boxes, although the loss is very low.

14

I can reduce my loss in both the training dataset and the test dataset to about 0.02, but my output of prediction visualization results is completely wrong, because there are...

chuxiang93

Issue with encoder.py

1

Would like to clarify the input for _get_anchor_boxes function under encoder.py is a tuple or a int? As I realised after the code went to line 50 it is not...

CluelessIT

你这也没个说明，然后也没有数据介绍，真的好难看懂，数据集总是提示不全

mayuelala

If an image has just background (no objects in entire image,so no need of bounding boxes) with class label 0, will this code give the correct classification as class 0...

sramakintel

Why do you freeze batch norm parameters when training?

4

would it be better to let batch norm parameters adapt to your current data?

leao1995

Update loss.py

7

Fix clc loss nan bug. Sum of cls_loss should divide total number of pos_neg

banjuanshua

Focal loss is very less

3

Hi, I have implemented your code and it worked properly but have the following concerns My sudo code works like this cls_targets = [batch_size, anchor_boxes, classes] # classes is 21...

prakashjayy

classification branch

2

@kuangliu HI For focal loss. The **classification branch** uses the **sigmoid function**. Why **background class** is considered in the classification branch?? such as **coco**, num_classes=80 instead of 81.

foralliance

investigate why the ap not good in voc dataset

4

I'v heard many achieved 60+ mAP only use voc0712-trainval for training, and I got similar result in my experiment. The mAP during training are shown in the figure. I'v achieved...

xytpai