Junhao Huang

@zfc929 Do you know why the confidence is "15.7315" rather a number between 0 and 1? Also the bounding boxes I got with last 4 number are nonsense, and many...

@zfc929 Yes, I used [prototxt](https://github.com/chenzhi1992/TensorRT-SSD/blob/master/ssd_deploy_iplugin.prototxt) provided by chenzhi1992, and official pre-trained caffemodel `VGG_VOC0712_SSD_300x300_iter_120000.caffemodel`. Interesting thing is that though bounding box locations does not make sense, many classifications (which is the...

@quocbh Thanks so much! I didn't notice that. It works! I can correctly detect now!

@quocbh I use opencv to draw ``` for (int k=0; k

@Optimus1072 I fork and push it. But in order to make it work, you'll still need modify some path, etc. @quocbh Not deploying on tx2 yet, but I got about...

> 1、判断两点之间是否有第三点的代码看懂了，但是这么做的原理是什么没想明白，虽然我验证了几组数据这么做是对的。代码在process.py的pass_through_third_point函数 原理是通过方向向量的点积大小判断是否同向。具体来说，设第一点，第二点，第三点为 $`P_1`$，$`P_2`$，$`P_3`$，如果$`\frac{P_1 P_3}{|P_1 P_3|} \cdot \frac{P_3 P_2}{|P_3 P_2|} > 0.8`$，表示向量 $`P_1 P_3`$ 和 $`P_3 P_2`$ 方向基本一致，说明 $`P_1 P_2`$ 经过 $`P_3`$。 > 2、struct.py的detemine_point_shape函数，我从代码里读的意思是根据角点倾斜角和两点的倾斜角差来判断角点的类型，但是config里的两个ANGLE_DIFF取值我有点不太懂，请问这块的原理依据是什么呢。 `BRIDGE_ANGLE_DIFF` 和 `SEPARATOR_ANGLE_DIFF` 是人工设置的两个阈值，分别用于判断 entrance-line 和...

这两个 diff 不是入口线和分割线的角度差，它们分别是用于判断入口线和分割线方向向量方向是否一致的阈值。 如果一个 predicted 入口线方向向量和 ground truth 入口线方向向量的夹角 < `BRIDGE_ANGLE_DIFF`，我们认为 predicted 入口线方向和 ground truth 入口线方向一致。 同理，如果一个 predicted 分割线方向向量和 ground truth 分割线方向向量的夹角 < `SEPARATOR_ANGLE_DIFF`，我们认为 predicted 分割线方向和 ground truth 分割线方向一致。

I think you are using the wrong directory for evaluation. Please use the `TEST_DIRECTORY` generated by `prepare_dataset.py`. https://github.com/Teoge/DMPR-PS/blob/master/README.md?plain=1#L68

If you have run the script `prepare_dataset.py`, the test directory should be located at `os.path.join(args.output_directory, 'test')` (https://github.com/Teoge/DMPR-PS/blob/master/prepare_dataset.py#L101).

我们提供的数据集 `label` 文件是包含 `marks` 和 `slots` key 的。 如果想要创建自己的数据集，请使用 [directional_point](https://github.com/Teoge/MarkToolForParkingLotPoint/tree/directional_point) 分支的`MarkToolForParkingLotPoint`标注工具进行标注，并使用 [mat2json.m](https://github.com/Teoge/MarkToolForParkingLotPoint/blob/directional_point/mat2json.m) 脚本生成 label 文件。