Fu comments

Results 11 comments of

Fu

training code

It took 3000ms in my conputer

This job is really great, the inference time for a point is about 10ms++, but the time for a full image is not much faster. Our GPU for the full...

Ask for DC model

missed dc model + 1

OneFormerProcessor

oneformer_ade20k_processor = OneFormerProcessor.from_pretrained("/xxx/HuggingFace/oneformer_ade20k_swin_large/")

intervl 微调后lora合并推理和直接用原来demo推理结果不一致

> do_sample=False 不然有随机性修改过了，但是还是不一致

多机多卡上是否可以正常训练

可以，不要你要自己修改实现，原始代码是不支持多机多卡训练的

Low Image Download Success Rate for GRIT Dataset Using img2dataset

@Vicky0812 , 可以分享一下您下载的50~60%的 grit的数据吗？我这里下载率更低

demo or code

thank you for your reply. 1. How many pseudo data were ultimately used for training? 2. With 16 Nvidia H100 GPUs, how many hours were used for training? 3. How...

demo or code

> Thank you for your attention. A demo will be provided in the coming days～😊 Looking forward to it.

demo or code

Is there a clear time point for demo ？ thank you