MagicSource

you should register custom dataset and train with custom data script.

can u try train coco first? for tiny dataset I think the lr is very hard to adjust. you can join our discord for further guidance.

根本原因是 多线程导致的 cpu 挂起（卡死），造成这个问题的原因很多，例如batchsize太大，内存小了等等。可以控制一下吞吐。

@curry8 怎么修改的啊

InferenceSample应该就是d2默认的配置，无法整除应该不是根本原因，很多数据及都可能出现无法整除，d2应该也不会出现这种低级bug 你可以按照 max_iter * image_per_batch 计算 总共计算了多少图片，与其他的配置epochs * coco_all_num 比对。d2默认是八卡的配置，理论上还得乘以 8/num_gpus.

Hello all, does the retargeting feature avaiable now?

@iamhankai Hi, I want extend tinynet on a object detection task, but the tinynet constructions coulpled with input resolution, how to take a suitable one for detection task like we...

No, I suggest using torchvision.ops.boxes.nms to do nms operation, however, which needs modification on detector. But since Alphapose already included that source code, it also worthy to modificate to standard...

Please using my fork version: it works on torch 1.11 https://github.com/jinfagang/AlphaPose

@vanlong96tg Have u train RFBNet with mobilenet? I trained about 2 days with 30 epochs, and the result got nothing, all the biggest probabilty are all background... can not detect...