Hi! How to perform test without Label_2 folder, like in KITTI testing?

Question

Hi! How to perform test without Label_2 folder, like in KITTI testing?

oljike opened this issue 5 years ago · 6 comments

Answer 1 · 2019-12-02T10:17:17.000Z

Hi, I have the same question, and am beginning to doubt how this model even works looking through the code

Answer 2 · 2019-12-02T14:57:06.000Z

Hi, for the test set, you can set the TEST.DATASET=test instead default setting val. For the test set without ground truth annotations, we only save the prediction results without evaluation.

Answer 3 · 2019-12-02T17:30:57.000Z

Hi, for the test set, you can set the TEST.DATASET=test instead default setting val. For the test set without ground truth annotations, we only save the prediction results without evaluation.

But in test dataset we don’t have the 2d rgb detection results .txt, we can not get the related frustum.

Answer 4 · 2019-12-02T19:20:57.000Z

I think the point is that we have no label_2 folder in testing to create the pickle files for TEST.DATASET, correct? In this case how do we make the pickle file to run inference on?

Answer 5 · 2019-12-04T14:36:13.000Z

@harish-kamath If you want to get the results of test set, you need to use a 2D detector to get the results of test set firstly, and save the results as the format in kitti/rgb_detections//rgb_detection_train.txt. And then use the kitti/prepare_data.py to prepare the pickle file( generate the pickle file from rgb_detection_val.txt). You need to change some of the code, including the path or image set file.

Answer 6 · 2019-12-04T14:38:23.000Z

@gujiaqivadin For the test set, we use the 2D detector provided by RRC for car category.