AllenCell / AllenCell/allencell-segmenter-ml

Catch-all notes only

未關閉
#284 0 則留言 0 個 reaction 已指派 1 人 已被 @thao-do 認領 在 GitHub 檢視
主要語言
Python
星號
5
分支
3
PR 合併指標
30 天內沒有已合併 PR

描述

Notes for future features but come up during current discussion:
#### Curation:
- Ability to skip images
- Ability to accept all images
- Image pre-adjustment or auto-adjustment
- random image selection
- Overwrite masks: if user add more polygons to the Mask layer then resave the mask, currently a message pops up to ask if they want to overwrite the saved mask - we're assuming user know that all polygons (old & newly created) on the mask layer will be saved and not just the ones they just added. Update the wording to clarify this.
- allow recurating the image dataset (same data, different sorting or masking)
** note: new data = new model
- in existing model: if curation has not started, everything is as default of "Start a new model" route; if curation has started and not complete: resume progress; if curation has started and completed: show end screen with feedback message to state that and direct to Training tab (and option to restart curation from scratch - reload directory and all)
- "Use this segmentation" dropdown idea: checkbox for each mask to turn on (Nope: Option: Seg1, Seg2, Exclude or Merge - Merge will trigger open the optional Mask(s) area (both mask?? but excluding might be used regardless of merging)

#### Prediction
- View output images from directory option: progress bar if prediction is going to take a while, showing the number of files being completed; when prediction on an image directory is completed, a popup dialog notifies the user, with an action button to view their images (loaded in set of raw/pred/original seg), also we can address the thresholding adjustment for probability output format (current format is binary with default threshold of 0.5) - this could potentially be a 4th module for image navigation in a folder, image adjustment, and image annotation (veryyyy lofty dream)
- Include a model LaminB1_tutorial

####Training
- allow continue training a model (all settings the same, just increase training step or time)
-

Specific UI tasks: Some might need to be split into smaller tasks or just more info:
- [ ] Update hover cursor when hovering over interact-able elements
- [ ] Consistent text styling for header etc. (add instances that need to be fixed here)
- [ ] Update button colors in Mask section >> higher priority than the rest
- [ ] Update button's states' styling (consistent style and color changed based on state)

貢獻指南

這個儲存庫沒有索引到貢獻指南

評估

這個 Issue 還沒有評估資料。

把新 issue 寄到你的電子郵件信箱

精選適合新手參與的 GitHub issue 摘要。