docs: Release version 1.1

CHANGES.md

@@ -0,0 +1,58 @@
+# Changelog
+
+## v1.1
+
+### New dataset version 1
+
+Before, Quevedo datasets were not versioned. Now, a field has been added to the
+`config.toml` file to track the Quevedo dataset schema, and a `migrate` command
+has been added to let users upgrade datasets to new versions.
+
+Some of the following additions are incompatible changes to dataset
+functionality and annotation files which have made this version upgrading
+necessary.
+
+### Tags are now a dictionary
+
+Grapheme tags (both free and bound) are now represented with a dictinary, with
+keys the names in the dataset `tag_schema`. This makes Annotation objects easier
+to use in custom code. The change affects both the library code and the files on
+disk, hence the migration.
+
+### Splitting now works differently
+
+Instead of assigning annotations to either a train or test split, they are
+assigned to a "fold". Groups of folds can then be defined as being available for
+train or for test (or none). This is also an incompatible change to annotation
+code and file representation. Old partitions will be lost, so after migration
+you will need to run the `split` command again.
+
+### Net configuration improvements
+
+- Detection networks can now have `width` and `height` parameters to tune
+    network input size.
+- All networks can now have a `max_batches` parameter to customize when to stop
+    training the net. This can serve to prevent overfitting and shorten training
+    times.
+
+### Annotation flags
+
+A new option "flags" has been added to the `config.toml` file. These flags are
+matadata values just like those in `meta_tags`, so assigned to both Logograms
+and Free Graphemes. The difference is that they are presented as checkboxes in
+the web interface, and shown as icons in annotation listings. This can serve to
+quickly mark annotations for annotators, for example if some have dubious or
+problematic tags, need some other kind of attention, or simply you want to keep
+track of them.
+
+### Other
+
+- When building the tag map for darknet, user tags are combined using the ASCII
+    FS character instead of "_", which can be problematic if tag values in the
+    dataset contain "_". This is an internal change and user code and data
+    should not be affected.
+- The `dataset.get_network` method now returns the same `Network` object if
+    called many times with the same network name. This helps save memory, which
+    in the case of neural networks can be crucial, without requiring the user to
+    keep the network in their own variable.
+- The web interface now can be used with touch on mobile devices.

ROADMAP.org

@@ -0,0 +1,37 @@
+#+title: Roadmap
+
+* Next
+
+** TODO Web interface improvements
+
+*** TODO Filter/search annotations in listing according to some tag(s).
+
+*** TODO Autosuggest values for tagging
+maybe use https://developer.mozilla.org/en-US/docs/Web/HTML/Element/datalist
+
+*** TODO Improve messaging
+Remove "saved" message when doing changes (inconsistent), maybe load indicator
+
+*** TODO Allow deleting entries
+Maybe just move the last to the hole?
+
+*** TODO Better mobile interface
+Maybe integrate with camera/scan app, make quevedo a target for "sharing"
+(uploading) images.
+
+*** TODO Web user improvements
+User groups, record annotator in json.
+
+*** TODO Documentation
+
+- [ ] "flags" (from v1.1)
+- [ ] changes from v1.2
+
+** TODO Improve use of scripts when modifing images
+
+When scripts modify images, don't save them, but return that it has been
+modified (ie return modified_tags, modified_img) and then it is `run_script`
+that saves the image to the appropriate path. Coversely, in the web interface
+the updated image can be sent to the frontend to be previewed, and if they want
+to save it send it back to the server on "save". The complication is that the
+image is now frontend state, not just a src link.