hunwenpinghao / ssd_convlstm_final

This is a consecutive frame target detection project by adding convlstm to SSD

SSD_ConvLSTM_final

This is a consecutive frame target detection project by adding convlstm to SSD.

Description

By adding conv LSTM to SSD, the highest accuracy is improved by nearly 10%. We have carried out on the radar data set provided by 25 customers. There are three targets in total. The network improves target 2 and target 3 greatly, target 2 increases by 5%, target 3 increases by 10%, and target 1 only improves by less than 1%.

Make DataSet

You can use the split_vidseq.py in ./data/ to make your own dataset. Since my data is confidential, it is not convenient to release it.

The difference between SSD and SSD_ConvLSTM:

SSD constructure:

SSD+ConvLSTM constructure:

ssd_convlstm_final's Issues

where can I download the dataset you used?

And I wish you tell me how to configure this dataset named "RDVOC_2019"?

请问RDVOC数据集是怎么制作的

可否发布制作RDVOC数据集的方法，或者您用的RDVOC数据集的格式，我好只做自己的数据集，非常感谢

输入的数据格式

请问你的输入数据是已经处理好的RD数据还是原始的ADC数据呀，我也是做RD目标检测的

可以给出最后数据文件结构的截图吗

你好，可以给出你最后数据的结构吗，就是数据有哪些文件夹，每个文件夹里有哪些内容，txt或者json文件里是怎么写的，截个图就行，根据这些我就能构建自己的数据集并进行训练了，现在看那个plit_vidseq.py 不是很清楚最后的数据结构是什么样的