search

Kemo-Huang / JMODT

Joint Multi-Object Detection and Tracking with Camera-LiDAR Fusion for Autonomous Driving
MIT License
96 stars 25 forks source link

Multiple GPUs for training #4

Open lx-ynu opened 3 years ago

lx-ynu commented 3 years ago

Hello, when I used multiple GPUs for training, it reported the following error, how should I solve it?

image

Bruce-Si commented 3 years ago

Hey, did you use a single GPU for training? Is there any problem?

zkp0113 commented 2 years ago

Excuse me，I meet same problems. No.1 ”UnboundLocalError: local variable 'val_loss_epoch' referenced before assignment“； For this problem I change some code then the problem fixed，but it comes new problem that same with this.

No.2 “best_model.path” is not update.

对不起，我也遇到同样的问题。 第1条问 “UnboundLocalError:赋值前引用的局部变量‘val_loss_epoch’”；对于这个问题，我更改了一些代码，然后问题得到了解决，但随之而来的是新问题。 第2条 “best_model.path”未更新。

Muhammadadeelaltaf commented 1 year ago

Hello everyone, how did you solve it? I am also facing the same errors while using 2 gpus with batch size 2: python3 tools/train.py --data_root data/KITTI --ckpt pretrained_model/jmodt.pth --finetune --batch_size 2 --output_dir output --mgpus cuda error cuda error 2