huangzicheng comments

Results 6 comments of


                                            huangzicheng

type(self).name, name)) AttributeError: 'YOLOv3' object has no attribute 'module'

I solve this problem , change it with: model_dict = dict({ 'epoch': epoch + 1, 'state_dict': model.state_dict(), }) torch.save(model_dict, os.path.join(args.save_dir, save_prefix+'_'+repr(epoch)+'.pth'))

Test results are bad，when I train 120 epoch, it class confidence are so low，is lower than 0.0001

![Detection_screenshot_11 03 2020](https://user-images.githubusercontent.com/30286349/76372264-cc6a5480-6377-11ea-8e39-c514c478e95e.png)

How can I train on my own dataset to detect new classes?

> I want to use CornerNet-Lite to detect new classes (not involved in COCO classes) on my own dataset, how can I do this? Or you can write it on...

RuntimeError: CUDA out of memory. Tried to allocate 22.00 MiB (GPU 0; 10.91 GiB total capacity; 10.03 GiB already allocated; 18.12 MiB free; 10.17 GiB reserved in total by PyTorch)

原因找到啦：不能直接pip install apex 卸载从源码安装 pip uninstall apex git clone https://www.github.com/nvidia/apex cd apex python setup.py install

你好，gc_prune.py里面对分组卷积后的model参数保存，可否提供下代码

Group =s , 分组操作使得卷积核为原来的channel/s, 代码里面的裁剪通道还是channel，比如（input=256,output=256, group=2）其实只有128个卷积通道，代码里面的out of range

i change detector ,use FCOS instead of YOlO, The effect of pedestrian detection is good, i want to Detect a frame to track 10 frames,it seem some trouble,can you show your code for real time processing

I want to speed up the real-time processing speed, detect a frame of about 5-10fps, track a frame of about 80fps, there are two ways of tracking, track_with_detect and track_without_detect,...