Chen Qibo

Hi, when I reproduced the data from the paper, I found that the data in your Table 5 differs from the KITTI leaderboard, is there any difference in the implementation...

Questions about pre-trained models

2

Hi, I used the your pre-trained model 'sdn_kitti_object.pth.', but the results of the test on train set are as follows： [2021-12-18 15:26:29 main.py:214] INFO TEST Epoch0:0 L 1.770 RLI 3.965...

Hi, I found that the code uses a fixed 0.54 baseline for data loading. Is there any difference between using a dynamic baseline and a fixed baseline?

When I use the SOLOv2_X101_DCN_3x pre training model and configuration in inference_demo.py, the display tells me that I need 64g memory, but I only have a 2080ti. Do you have...

My download speed is really slow

NaN values in the data

6

Hello, I see that Nan exists in the data in the FAQ of the website of the dataset. Do I want to know whether the Nan value exists in disparity...

 Can I just use Pytorch?

What is the number of bits of image data provided in your dataset?

3

Is it 8bit or higher

训练显存问题

3

请问一下你们在用mmyolo和mmdet训练yolo world模型的时候有没有遇到显存增长的问题。我使用mmdet中的rtmdet构建了类似方案，但训练时存在显存泄露，

我在使用RTMDet训练类似O365v1这样的大型数据集时，发现显存占用远超COCO数据集，经过torch.cuda.max_memory_allocated()定位显存激增的位置发生在DynamicSoftLabelAssigner中的 soft_cls_cost = F.binary_cross_entropy_with_logits( valid_pred_scores, soft_label, reduction='none') 目前我不清楚，为什么会导致显存激增，但我使用gpu_assign_thr暂时缓解了这个问题，有相同经历的朋友可以参考[PR](https://github.com/open-mmlab/mmdetection/pull/11467)，也希望有懂得大佬解答一下。 我的环境是: torch 2.1.0 numpy 1.23.0 mmdet 3.3.0 mmcv 2.1.0 mmengine 0.10.3