Dataset format about keras-image-segmentation

[dhkim0225]

Download these files!!

leftImg8bit_trainvaltest.zip (11GB)
https://www.cityscapes-dataset.com/file-handling/?packageID=3

gtFine_trainvaltest.zip (241MB)
https://www.cityscapes-dataset.com/file-handling/?packageID=1

[dhkim0225]

In fact, I did not understand exactly what you were asking. I will write some more information.

We were able to parse the dataset into numpy and create the .h5 file. we could get the data from data.h5 about 10 times faster than opencv imread.

When creating the data, the image was resized(256, 512, 3) and limited to three classes.

[dhkim0225]

If you want to train your own dataset,,,,, Sorry but its not supported now. We`ll make methods later(maybe after several months - maybe after we make various sota models)

[tkwoo]

I recommend you to use original cityscape dataset structure thuogh we use hdf5 format.

cityscpae dataset sturcture

The folder structure of the Cityscapes dataset is as follows:

{root}/{type}{video}/{split}/{city}/{city}_{seq:0>6}_{frame:0>6}_{type}{ext}

The meaning of the individual elements is:

• root the root folder of the Cityscapes dataset. Many of our scripts check if an environment variable CITYSCAPES_DATASET pointing to this folder exists and use this as the default choice.
• type the type/modality of data, e.g. gtFine for fine ground truth, or leftImg8bit for left 8-bit images.
• split the split, i.e. train/val/test/train_extra/demoVideo. Note that not all kinds of data exist for all splits. Thus, do not be surprised to occasionally find empty folders.
• city the city in which this part of the dataset was recorded.
• seq the sequence number using 6 digits.
• frame the frame number using 6 digits. Note that in some cities very few, albeit very long sequences were recorded, while in some cities many short sequences were recorded, of which only the 19th frame is annotated.
• ext the extension of the file and optionally a suffix, e.g. _polygons.json for ground truth files

Possible values of type

• gtFine the fine annotations, 2975 training, 500 validation, and 1525 testing. This type of annotations is used for validation, testing, and optionally for training. Annotations are encoded using json files containing the individual polygons. Additionally, we provide png images, where pixel values encode labels. Please refer to helpers/labels.py and the scripts in preparation for details.
• gtCoarse the coarse annotations, available for all training and validation images and for another set of 19998 training images (train_extra). These annotations can be used for training, either together with gtFine or alone in a weakly supervised setup.
• leftImg8bit the left images in 8-bit LDR format. These are the standard annotated images.
• leftImg16bit the left images in 16-bit HDR format. These images offer 16 bits per pixel of color depth and contain more information, especially in very dark or bright parts of the scene. Warning: The images are stored as 16-bit pngs, which is non-standard and not supported by all libraries.
• rightImg8bit the right stereo views in 8-bit LDR format.
• rightImg16bit the right stereo views in 16-bit HDR format.
• timestamp the time of recording in ns. The first frame of each sequence always has a timestamp of 0.
• disparity precomputed disparity depth maps. To obtain the disparity values, compute for each pixel p with p > 0: d = ( float(p) - 1. ) / 256., while a value p = 0 is an invalid measurement. Warning: the images are stored as 16-bit pngs, which is non-standard and not supported by all libraries.
• camera internal and external camera calibration. For details, please refer to csCalibration.pdf
• vehicle vehicle odometry, GPS coordinates, and outside temperature. For details, please refer to csCalibration.pdf