Not able to split the kitti dataset about 3d_detection

[m-parchami]

If by 'splitting' you mean splitting by space in the code, make sure you are working with the right dataset. There are 2 annotation files that you can use for doing this 3D detection. Here and here. The former is called "raw" not in the sense that it hasn't been annotated; instead, it has all of the data that can be used for many purposes. For example, there are cars that are completely occluded by another car in the front, yet they are labeled for different purposes. These files can be used for having the point clouds data as well. Reading from the raw dataset can be challenging if you decide to completely do it by yourself. Make sure you use the open-source tools.

In contrast, you can use the second link for much simpler use cases. Just some images, with patches and basic information about the vehicle in the patch (e.g. rotation, dims, translations, type, etc.). The values are given line by line and are MUCH simpler to parse. This code uses this dataset as well. As you can see here at the beginning of the parse_annotation function in process_data.py:

for line in open(label_dir + label_file).readlines():
            line = line.strip().split(' ')
            truncated = np.abs(float(line[1]))
            occluded = np.abs(float(line[2]))
            cls = line[0]

            # add objects to train data only when their truncated level <0.3 and occluded level <= 1
            if cls in cls_to_ind.keys() and truncated < 0.3 and occluded <= 1:
                theta_loc = -float(line[3]) + 3*np.pi / 2.
                # Make sure object's theta_loc is in [0..2*pi].
                theta_loc = theta_loc - np.floor(theta_loc / (2. * np.pi)) * (2. * np.pi)

                obj = {'name': cls,
                       'image': image_file,
                       'xmin': int(float(line[4])),
                       'ymin': int(float(line[5])),
                       'xmax': int(float(line[6])),
                       'ymax': int(float(line[7])),
                       'dims': np.array([float(number) for number in line[8:11]]),
                       'theta_loc': theta_loc
                       }

The first link has XML files and the second has naive .txt files that you can easily split by space.

So, check your dataset ;)

[pankaj251003]

Thank you so much for the response ......i have downloaded the left color image dataset from kitti and trying to split training dataset in train and test as per the train.txt and test.txt given, it gives the deep copy error, also it does not test the other images from my pc. please help @m-parchami

[m-parchami]

Ok. Please share the splitting code and the error so that you can receive some help.

[pankaj251003]

thank you ...i am asking how to use testing dataset of kitti in this model .....now i am able to train the model with training dataset which has 7481 images with label file...........please tell me how to use testing dataset of kitti