fix and update Caltech101 target types #7752
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/7752
Note: Links to docs will display an error until the docs builds have been completed. ❗ 1 Active SEVsThere are 1 currently active SEVs. If your PR is affected, please view them below: ❌ 1 New Failure, 2 Unrelated FailuresAs of commit 2e4e39d: NEW FAILURE - The following job has failed:
BROKEN TRUNK - The following jobs failed but were present on the merge base cc0f9d0:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
| ``annotation``. Can also be a list to output a tuple with all specified | ||
| target types. ``category`` represents the target class, and | ||
| ``annotation`` is a list of points from a hand-generated outline. | ||
| target_type (string or list, optional): Type of target to use, ``category``, ``annotation``, ``box_coord``, or |
There was a problem hiding this comment.
I can't really think of a good reason to use box_coord instead of bbox. Don't we use bbox everywhere else?
I know we are "using what the original dataset defines" but that's.... probably a bad idea? Doesn't it just lead to more and more inconsistencies? We don't have to keep doing this right?
| - ``box_coord`` (np.ndarray, shape=(1, 4), dtype=int): Bounding box in XYXY format | ||
| - ``obj_contours`` (np.ndarray, shape=(N, 2) dtype=float): Contour vertices in XY format |
There was a problem hiding this comment.
Do we really need to support individual box_coord and obj_contours? Can't we let users use the keys that they want in the dict?
Regardless, we really want the format to be consistent. Let's just use XYXY everywhere instead of leaving the annotation "raw".
Fixes #7748. This PR reflects my proposal in #7748 (comment).
ToDo