CenterNet

CenterNet copied to clipboard

画出检测结果之后，同一个目标对应好几个类别标签，这个问题该怎么处理呢

lalalagogogochong

Integrating EfficientNet with CenterNet

5

I am trying to integrate efficientnet from https://github.com/narumiruna/efficientnet-pytorch/blob/master/efficientnet/models/efficientnet.py. I added efficientnet.py in models/networks folder and added corresponding code for factory, etc. Then, I tried to train the network using COCO...

chwong1

shenyi0220

Lower the weight of offset loss has already done.

yyhhlancelot

The train loss not change

6

Thanks to your great job. When I train the network use command: python main.py ctdet --exp_id coco_dla_2x --batch_size 128 --lr 1.25e-4 --gpus 0 --num_workers 0 --val_intervals 5 --num_epochs 100 --master_batch...

trustguan

When the input height and input width are both 512, the results are normal, which seting the input width 512 and input height 320, the output is abnormal

ShihuaiXu

error in modulated_deformable_im2col_cuda: no kernel image is available for execution on the device

3

i ran train in google colab and i have a Issues error in modulated_deformable_im2col_cuda: no kernel image is available for execution on the device input: %cd /content/gdrive/MyDrive/CenterNet/src !python main.py ctdet...

knotgrass

Does CenterNet support pose/keypoint estimation for more than 1 object category at the same time?

3

Suppose I have two object categories with completely different sets of keypoints (e.g. human keypoints and cat keypoints). Can I train CenterNet to detect keypoints for both categories in one...

jackkwok

Not sure how to change annotation path or category information

1

Hi guys. This is my first machine learning project and I'm excited! I'm also lost because all of this is very new to me. I managed to get past the...

howdidthatwork

关于在自定义数据集上训练的一些问题？

9

@xingyizhou 大佬你好，就是我是一名大二学生，然后是在中北大学的robomaster战队里负责用神经网络识别装甲板实现自动瞄准，不过就是之前我用yolo系列训练出来的模型最后实际测试时得到的bbox和装甲板的轮廓并不能很好的拟合，导致后续使用pnp进行姿态解算时会有较大误差，所以我想将传统yolo的数据集格式改为用四个角点的坐标，然后我现在的数据集格式是像这样：1 0.673029 0.373564 0.678429 0.426232 0.830433 0.401262 0.824525 0.351212，第一个数字是类别id，后面八个数字是归一化后的四个角点坐标，我现在就是想用centernet来训练，我看了您仓库中给的在自定义数据集上训练的博客，我现在知道了应该是数据集应该是coco格式的，然后我就想问一下我可以直接用cvat这类工具标注后导出为coco格式然后来训练吗，这样的话我标注的时候可以使用四点标注的bbox吗？望大佬解惑

Hezhexi2002