docs: Classes vignette - #404
srishtiii28 wants to merge 4 commits into
Conversation
cregouby
left a comment
cregouby
left a comment
There was a problem hiding this comment.
Praise It is a very clear explanation of the internals of each and every S3 class, with SVG well structured, including ARIA markup (thanks for that ! )
todo 'Visualization utilities' lives at articles/examples/visualization-utilities.html — reachable from the Examples menu but not from the Articles menu added in this PR. suggestion Moving 'Visualization utilities' vignette into the Articles menu would make the relationship explicit.
| | family | dispatches on | acts on | | ||
| |---|---|---| | ||
| | `transform_*()` | `torch_tensor`, `magick-image`, `array`, `matrix` | the image alone | | ||
| | `target_transform_*()` | `object_detection_target`, `segmentation_target` | the target alone | |
There was a problem hiding this comment.
todo misleading The table entry implies that all target_transform_*() functions accept both classes. They do not. The dispatch groups are strictly disjoint:
- target_transform_resize, target_transform_rotate, target_transform_affine, target_transform_sahi_crop — object_detection_target only
- target_transform_coco_masks, target_transform_trimap_masks — segmentation_target only
A user who calls target_transform_rotate() on a segmentation_target will get a "no applicable method" error. The table as written directly contradicts that behavior.
suggestion Split the row into two, or add a qualifier such as "detection target transforms" vs "mask-building transforms."
| Model outputs are plain lists and carry no class of their own. The object | ||
| detection models return `$detections`, one element per image of the batch, each | ||
| with `$boxes`, `$labels` and `$scores`; the semantic segmentation models return | ||
| `$out`, the per-class logits. The "Visualization utilities" article covers |
There was a problem hiding this comment.
todo 'Visualization utilities' is not in "Article" and should be turned into a link.
suggestion move the 'visualization utilities' files in the created 'Article' menu.
There was a problem hiding this comment.
todo missing A Small context introduction of the why we introduced S3 classes is a must.
todo missing As those class are a kind of coverage of torch / vision TVTensors, a trailing paragraph at the end of the vignette for "Comparison with torch / vision TVTensors (advanced)" is worth adding, highlighting the differences for people used to manipulate TVTensors
This branch has not been deployed
2 participants
Closes #396.
Adds a vignette that goes through the S3 classes a dataset item can carry and what is inside each one.