Tables
Acceptance Criteria
DELIVERABLES | ACCEPTANCE Criteria |
---|---|
Label file |
A XML / CSV / JSON file with the labeled meta data in format decided by mutual discussion with VW or specified by VW |
Annotated and Tagged images |
SEMANTIC PIXEL LABELING: 1. At least 97% of the total pixels in the camera image are correctly annotated 2. At least 97% of the pixels of each individual an object are correctly annotated 3. Maximum distance between object contour in image and annotated object contour BOUNDING BOXES: 1. Complete annotation of all objects and attributes according to class definition 2. Unique IDs for tracked objects for sequences 3. 2D and 3D bounding boxes in the camera image: maximum distance between bounding box and object contour: maximum 3 pixels and maximum 3% of the bounding box width or height 4. 3D Bounding Boxes in point clouds: Bounding Box bounding surfaces are congruent with the object bounding points in the point cloud |
Subjective labeling of attributes |
This peculiar case of subjective labeling results shall be accepted by the mutual discussion |
Temporal annotation |
The tolerance of ± 2 frames (200 msec) is accepted on both the start and stop time stamps |
Productivity Report in terms of: • Manual Rate • Automation + Manual Rate |
A report showcasing % of work processed: - % of total data processed manually. - % of total data processed with the combination of Automation & Manual Effort. |
Process for the review |
At every periodic milestone, there will be a feedback about the annotation and correction required in the same. There will be a revision check and correction phase from KPIT to improve output as per feedback. This constant interaction is expected to ensure as well as sustain the quality |