yook

A question about detection results which in data/KITTI/detection/

9

for example,the detection results in data/KITTI/detection/pointrcnn_Car_val as follow: 0,2,298.3125,165.1800,458.2292,293.4391,8.2981,1.9605,1.8137,4.7549,-4.5720,1.8435,13.5308,-2.1125,-1.7867 0,2,1050.4751,177.0771,1241.0000,239.3750,1.3104,1.4905,1.6289,4.0662,13.6684,1.6046,18.4125,-0.1005,-0.7391 0,2,364.7274,137.8874,453.8243,172.5940,-0.4501,1.4779,1.5952,3.8674,-9.0942,-0.0122,32.8211,-1.0285,-0.7582 1,2,295.6248,166.5919,452.0338,291.1647,9.1349,1.9552,1.7492,4.3544,-4.6583,1.8572,13.6244,-2.1461,-1.8166 1,2,356.1560,159.8431,391.7758,178.2208,1.4258,1.4772,1.5754,3.9679,-19.6023,0.4317,59.9970,1.5974,1.9132 1,2,139.4961,190.2121,223.3792,223.3236,0.7900,1.4796,1.5794,3.7088,-21.4827,2.3998,36.2672,1.7515,2.2863 2,2,294.6839,155.8907,449.7736,279.9171,8.8634,1.9858,1.7958,4.3856,-4.7726,1.7146,13.8804,-2.1411,-1.8100 I found that the 2D bbox, score, and alpha are different from the Kitti dataset....

不是要损失、学习率曲线，而是需要每一轮的训练情况，类似下面这样的： ----------- AP40 Results ------------ Pedestrian [email protected], 0.50, 0.50: bbox AP40:37.4608, 33.8643, 32.5023 bev AP40:37.5405, 33.7067, 31.7938 3d AP40:30.3489, 27.1389, 24.6025 aos AP40:15.51, 14.54, 14.02 Pedestrian [email protected], 0.25, 0.25: bbox...

The code in the original paper is too difficult to understand, and I cannot summarize the changes made in the code compared to the original text.

请问这个项目是检测KITTI所有类别吗？

4

我在融入跟踪算法的时候发现多了很多类别，想确认一下。

map仅仅只有0.0003

7

我使用mini数据集进行了训练，在第22个epoch的时候，程序卡死，然后我对最后的两个epoch进行了验证，结果map出奇的低。 ================epoch 20================= AP: 0.0003 mATE: 1.0000 mASE: 1.0000 mAOE: 1.0000 mAVE: 1.0000 mAAE: 1.0000 NDS: 0.0001 Eval time: 0.4s Per-class results: Object Class AP ATE ASE AOE AVE AAE...

我使用nuscenes-mini训练了20个epoch，但是得到了较低的map

9

这是因为mini数据集原因吗？有没有训练好的模型提供呢？ Filtering predictions => Original number of boxes: 12177 => After distance based filtering: 9922 => After LIDAR points based filtering: 9922 => After bike rack filtering: 9915 Filtering ground...

**环境是：** - open3d 0.15.2 - python 3.8.18 - pytorch 1.10.0(cuda11.3,cudnn8.2) - pytorch-lightning 1.5.10 - pytorch3d 0.6.2 - shapely 1.7.1 - torchvision 0.11.0+cu113 **仅仅因为显存原因调整了batch_size为8，训练时异常：** > model.backward(closure_loss, optimizer, *args, **kwargs) > File...