search

lyuwenyu / RT-DETR

[CVPR 2024] Official RT-DETR (RTDETR paddle pytorch), Real-Time DEtection TRansformer, DETRs Beat YOLOs on Real-time Object Detection. 🔥 🔥 🔥
Apache License 2.0
1.65k stars 179 forks source link

Train problem with paddle #238

Open chenhao324 opened 3 months ago

chenhao324 commented 3 months ago

Why is it that when I run the program to the first convolution operation of the hgnet_v2. py file, there is no progress in the program, and the graphics card memory usage is relatively low (about 2GB), and there are no errors reported at this time 1711100553906

lyuwenyu commented 3 months ago

没太看懂你啥问题

chenhao324 commented 3 months ago

Hello author, thank you for reading my question. My question is that when I was training a model using the PaddlePaddle framework, the program stopped executing when it reached the first convolution calculation, and there was no response. I waited for a long time without making any progress

lyuwenyu commented 3 months ago

What data are you using? I haven't encountered this problem using mscoco data.

ps. my paddle version is

paddlepaddle-gpu==2.4.2
# or 
paddlepaddle-gpu==2.5.0
chenhao324 commented 3 months ago

I am using imagenetVID data, which is converted to Coco format and can run successfully on the PyTorch framework. However, this issue occurs on the Paddle framework. To match my cuda10.1 environment, I am using PaddlePaddle=2.3.2

lyuwenyu commented 3 months ago

please upgrade paddlepaddle and try again.