fuweifu-vtoo comments

Results 9 comments of


                                            fuweifu-vtoo

I download the datasets from Official website，but it doesn't have 'labelTrainIds.png', i want to know what's the difference between 'labelTrainIds.png' and 'labelIds.png'

I have the same question, and I noticed that you have "line = line.replace('_leftImg8bit', '_gtFine_labelTrainIds')"this code in your utils.py, can i change this and get the same answer?

why not support Multi-GPU? is it not necessary for UDA?

It feels like your code supports multi-gpu training natively because you use the DDP wrapper. You just need to start the command with OMP_NUM_THREADS=1 python3 -m torch.distributed.launch ?

When the target to be detected is very small, the loss is found to be almost 0 during the training process.

When I detect larger objects, everything is normal, including loss and results. So my environment and code are all fine.

您好

链接：https://pan.baidu.com/s/1XtSKgjYlAgsECKXX69tpdA?pwd=otb3 提取码：otb3

Memory issue

when batch size=2, around 29G with fdthing=True, around 15G with fdthing=False

About the T-Rex / T-Rex2 demo

why not implement T-Rex2 along the lines of grounding DINO

Thank you for your detailed explanation～ Another question I also hope to get your answer is, does T-Rex2 freeze the CLIP text encoder during training?

why not implement T-Rex2 along the lines of grounding DINO

Also, how long does it take to train a T-Rex2 with swin Transformer tiny model on 16 NVIDIA A100 GPUs with a total batch size of 128?

why not implement T-Rex2 along the lines of grounding DINO

Hi, how long did the first phase of training take?