engineer, how should the data for multitask be prepared, and what information is placed in each .txt file?

[zcstart]

engineer, how should the data for multitask be prepared, and what information is placed in each .txt file?

for example kitti def _init(self): for line in open(osp.join(self.root, self.image_set + '.txt')): print('---- line ----') print(line) img_path, label_path, focal_length, baseline = line.split(' ')

line should contain contents?

[wangxd-xlnx]

Hi @zcstart

Could you describe which vitis-ai version of which model you are using？In vitis-ai 3.5, model zoo does not have this model. Or you mean the examples of vitis-ai library?

[zcstart]

i use vitis-ai 2.5 but dataset i do not understand how to prepare data

[zcstart]

01 cityexacpe get 16 class 。which 16 class? 02 depth dataset。 how to understand Depth estimation test images: kitti eigen split? 03 lane dataset。1) data\multi_task_det5_seg16\lane\train\images 2)data\multi_task_det5_seg16\lane\train\seg 04 drivable dataset。data\multi_task_det5_seg16\drivable\train\images 2)data\multi_task_det5_seg16\drivable\train\seg

[wangxd-xlnx]

Hi @zcstart 1.The segmentation dataset is CityScapes, and the original data has a total of 19 categories.
The specific 16 categories need further confirmation and will feedback later. 2.The depth dataset contains data meaning img_path : Training RGB image path label_path : label path focal_length : focal length baseline : the distance between the two cameras

3. lane dataset data\multi_task_det5_seg16\lane\train\images: training data RGB images data\multi_task_det5_seg16\lane\train\seg: The label for the RGB images of the training data
4. drivable dataset Contains classes: 'car', 'sign', 'road' data\multi_task_det5_seg16\drivable\train\images: training data RGB images data\multi_task_det5_seg16\drivable\train\seg: The label for the RGB images of the training data

[zyh321]

Hi @zcstart 1.The segmentation dataset is CityScapes, and the original data has a total of 19 categories. The specific 16 categories need further confirmation and will feedback later. 2.The depth dataset contains data meaning img_path : Training RGB image path label_path : label path focal_length : focal length baseline : the distance between the two cameras 3. lane dataset data\multi_task_det5_seg16\lane\train\images: training data RGB images data\multi_task_det5_seg16\lane\train\seg: The label for the RGB images of the training data 4. drivable dataset Contains classes: 'car', 'sign', 'road' data\multi_task_det5_seg16\drivable\train\images: training data RGB images data\multi_task_det5_seg16\drivable\train\seg: The label for the RGB images of the training data

Hi wangxd-xlnx About the segmentation dataset from 19 categories to 16 categories, how does this happen? Is there any files to process the original CityScapes?