webber.cai
webber.cai
> > > > > > > The data preparation is almost the same as MMDetection. You can refer [custom.py](https://github.com/jbwang1997/OBBDetection/blob/master/mmdet/datasets/custom.py) for the data structure. > > > > > >...
I also encountered this problem. I solved it by replacing false with ”false“.
hey, i also have this problem, the function “decodeDetections” is very very slow. almost 80ms in jetson nx.
same issue
I encountered a similar problem lmdeploy - ERROR - engine.py:950 - Task failed Traceback (most recent call last): File "llm_CoT/lmdeploy/lmdeploy/pytorch/engine/engine.py", line 945, in __task_callback task.result() File "llm_CoT/lmdeploy/lmdeploy/pytorch/engine/engine.py", line 998, in...
> > 1. 你重新混合数据训练的话,不是用 resume,而是应该用 load_from > > 2. 不管是啥数据都支持微调的,ODVG 的微调我们也有例子的,所以是可以的,可以看 LVIS 微调,那个就是 ODVG 格式微调 > > 3. 上面做法没有问题的,就是可能有些参数要调,不同 stage 和不同规模的数据,学习率啥的要调一下 > > 您好, 目前,我已经构建了训练的全流程,步骤如下: 1、使用少量的数据66张(od格式)进行微调,已经完成,检出率和置信度都很好,20个epoch,学习率为0.0001; 2、使用微调后的权重对数据集中其他未打标签数据生成伪标签(200张图片,od格式),设置置信度为0.7; 3、将第二步生成的伪标签和第一步的数据进行混合,并load_from第一步生成的微调权重,其余训练参数与第一步相同,进行微调训练。 > >...