# pyxelstitch
> Simple, matplotlib based tool for hand-labeling the two-image correspondences
## Install
`pip install pixelstitch`
## How to use
Let's test our annotator on a sample project. It needs a list of triplets: (`path_to_img1`, `path_to_img2`, `path_to_corrs_to_save`).
```python
import os
rootdir = 'sample_project'
pairs = os.listdir(rootdir)
img_pairs_list = []
for p in pairs:
if p == '.DS_Store':
continue
cur_dir = os.path.join(rootdir, p)
img_pairs_list.append((os.path.join(cur_dir, '01.png'),
os.path.join(cur_dir, '02.png'),
os.path.join(cur_dir, 'corrs.txt')))
print (img_pairs_list)
```
[('sample_project/ministry/01.png', 'sample_project/ministry/02.png', 'sample_project/ministry/corrs.txt'), ('sample_project/petrzin/01.png', 'sample_project/petrzin/02.png', 'sample_project/petrzin/corrs.txt')]
/opt/homebrew/Caskroom/miniforge/base/envs/python39/lib/python3.9/site-packages/ipykernel/ipkernel.py:283: DeprecationWarning: `should_run_async` will not call `transform_cell` automatically in the future. Please pass the result to `transformed_cell` argument and any exception that happen during thetransform in `preprocessing_exc_tuple` in IPython 7.17 and above.
and should_run_async(code)
Now we are ready to initialize `CorrespondenceAnnotator`. Don't forget to declare magic command ```%matplotlib widget```.
**WITHOUT MAGIC IT WOULD NOT WORK**
You also should explicitly specify, if you want to save (and possibly over-write previous better annotation) current correspondences automatically when clicking on **prev** and **next** buttons for going to the next pair.
```python
%matplotlib widget
from pixelstitch.core import *
CA = CorrespondenceAnnotator(img_pairs_list, save_on_next=True)
```
/opt/homebrew/Caskroom/miniforge/base/envs/python39/lib/python3.9/site-packages/ipykernel/ipkernel.py:283: DeprecationWarning: `should_run_async` will not call `transform_cell` automatically in the future. Please pass the result to `transformed_cell` argument and any exception that happen during thetransform in `preprocessing_exc_tuple` in IPython 7.17 and above.
and should_run_async(code)
Now we can run the annotation.
**Left-click** on the image to add a point
**right-click** -- to remove the point from both images.
### Matplotlib shortcuts:
- **o** for zoom
- **p** for pan (move)
It is also recommended to set full page width for the jupyter
```python
%matplotlib widget
from IPython.core.display import display, HTML
display(HTML("<style>.container { width:95% !important; }</style>"))
CA.start(figsize=(12,7))
```
# Controls
## Selectors
- `Model`. One can select between "`F`" -- fundamental matrix and "`H`" -- homography. The selection influences the reprojection error type, and the visualization of the models and reprojection errors, shown when clicked on `NextCorrsValPoint`, `NextCorrsValAll` and `ShowModel` buttons.
## Buttons
- `NextCorrsValPoint`. Shows the correspondence in the bottom axis. The image title shows correspondence index and the reprojection error. If `Model` is `F`, it will show induced epipolar line, if `H` -- the position of the reprojected point from other image. The model is estimated with all other correspondences except current one.
- `NextCorrsValAll`. Shows the correspondences in the bottom axis. Similar to `NextCorrsValPoint` button, but shows all points. The model is estimated with all correspondences except current one, which index is shown in the title.
- `ShowModel`. Has different behavoir depending on the `Model` selected. For `F` -- shows correspondences with their induced epipolar lines. Unlike `NextCorrsValPoint` and `NextCorrsValPoint` all correspondences are used for model estimation.
- For `H`, the button shows overlay of image 1 reprojected into image2 with image2. The reprojected area is defined by the convex hull of the labelled correspondences. Next click flips the order, i.e. shows the image 2 reprojected into image 1.
- `Save points` -- saves (overwrites) the correspondences into the text file.
- `Prev` -- Loads and shows previous image pair to label. If the `CorrespondenceAnnotator` was initialized with `save_on_next=True`, the current pair correspondences are saved before the switch. Change is not cyclical, so the button does nothing on 1st image pair
- `Next` -- Loads and shows next image pair to label. If the `CorrespondenceAnnotator` was initialized with `save_on_next=True`, the current pair correspondences are saved before the switch.Change is not cyclical, so the button does nothing on last image pair.
- `CLAHE` -- Images are shown with enhanced contrast with [CLAHE](http://amroamroamro.github.io/mexopencv/opencv/clahe_demo_gui.html) algorithm.
- `Pick 4 points for homography` -- Special mode. User picks 4 points in one image, which define new fronto-parallel view. This mode helps for labeling obscure views, see example below. The order of the points: top-left -> top-right -> bottom right -> bottom left. After the 4 point picked, the mode is switched off, so user needs to click the button again if she wants to rectify another image.
- `ResetView` -- Resets any recifications or zoom done to images.
### Rectification picking mode example
3 points are selected
All points are selected and image1 is rectified to the rectangle, defined by the selected points
Raw data
{
"_id": null,
"home_page": "https://github.com/ducha-aiki/pixelstitch/tree/master/",
"name": "pixelstitch",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.8",
"maintainer_email": null,
"keywords": "local features, correspondences, RANSAC, image matching, WBS, WxBS",
"author": "Dmytro Mishkin",
"author_email": "ducha.aiki@gmail.com",
"download_url": "https://files.pythonhosted.org/packages/d0/62/1e40420b1eeb4f2c4c8cbecd4d41ddffdfbac7275c56a46c91bc6aef1962/pixelstitch-0.1.5.tar.gz",
"platform": null,
"description": "# pyxelstitch\n> Simple, matplotlib based tool for hand-labeling the two-image correspondences\n\n\n## Install\n\n`pip install pixelstitch`\n\n## How to use\n\nLet's test our annotator on a sample project. It needs a list of triplets: (`path_to_img1`, `path_to_img2`, `path_to_corrs_to_save`).\n\n```python\nimport os\nrootdir = 'sample_project'\npairs = os.listdir(rootdir)\nimg_pairs_list = []\nfor p in pairs:\n if p == '.DS_Store':\n continue\n cur_dir = os.path.join(rootdir, p)\n img_pairs_list.append((os.path.join(cur_dir, '01.png'),\n os.path.join(cur_dir, '02.png'),\n os.path.join(cur_dir, 'corrs.txt')))\n\nprint (img_pairs_list)\n```\n\n [('sample_project/ministry/01.png', 'sample_project/ministry/02.png', 'sample_project/ministry/corrs.txt'), ('sample_project/petrzin/01.png', 'sample_project/petrzin/02.png', 'sample_project/petrzin/corrs.txt')]\n\n\n /opt/homebrew/Caskroom/miniforge/base/envs/python39/lib/python3.9/site-packages/ipykernel/ipkernel.py:283: DeprecationWarning: `should_run_async` will not call `transform_cell` automatically in the future. Please pass the result to `transformed_cell` argument and any exception that happen during thetransform in `preprocessing_exc_tuple` in IPython 7.17 and above.\n and should_run_async(code)\n\n\nNow we are ready to initialize `CorrespondenceAnnotator`. Don't forget to declare magic command ```%matplotlib widget```.\n**WITHOUT MAGIC IT WOULD NOT WORK**\n\nYou also should explicitly specify, if you want to save (and possibly over-write previous better annotation) current correspondences automatically when clicking on **prev** and **next** buttons for going to the next pair. \n\n```python\n%matplotlib widget\nfrom pixelstitch.core import *\nCA = CorrespondenceAnnotator(img_pairs_list, save_on_next=True)\n```\n\n /opt/homebrew/Caskroom/miniforge/base/envs/python39/lib/python3.9/site-packages/ipykernel/ipkernel.py:283: DeprecationWarning: `should_run_async` will not call `transform_cell` automatically in the future. Please pass the result to `transformed_cell` argument and any exception that happen during thetransform in `preprocessing_exc_tuple` in IPython 7.17 and above.\n and should_run_async(code)\n\n\nNow we can run the annotation. \n\n**Left-click** on the image to add a point \n\n**right-click** -- to remove the point from both images. \n\n### Matplotlib shortcuts:\n\n- **o** for zoom \n- **p** for pan (move)\n\nIt is also recommended to set full page width for the jupyter\n\n\n```python\n%matplotlib widget\nfrom IPython.core.display import display, HTML\ndisplay(HTML(\"<style>.container { width:95% !important; }</style>\"))\nCA.start(figsize=(12,7))\n```\n\n\n\n# Controls\n\n## Selectors\n\n- `Model`. One can select between \"`F`\" -- fundamental matrix and \"`H`\" -- homography. The selection influences the reprojection error type, and the visualization of the models and reprojection errors, shown when clicked on `NextCorrsValPoint`, `NextCorrsValAll` and `ShowModel` buttons.\n\n## Buttons\n\n- `NextCorrsValPoint`. Shows the correspondence in the bottom axis. The image title shows correspondence index and the reprojection error. If `Model` is `F`, it will show induced epipolar line, if `H` -- the position of the reprojected point from other image. The model is estimated with all other correspondences except current one. \n\n- `NextCorrsValAll`. Shows the correspondences in the bottom axis. Similar to `NextCorrsValPoint` button, but shows all points. The model is estimated with all correspondences except current one, which index is shown in the title. \n\n\n\n- `ShowModel`. Has different behavoir depending on the `Model` selected. For `F` -- shows correspondences with their induced epipolar lines. Unlike `NextCorrsValPoint` and `NextCorrsValPoint` all correspondences are used for model estimation. \n\n\n- For `H`, the button shows overlay of image 1 reprojected into image2 with image2. The reprojected area is defined by the convex hull of the labelled correspondences. Next click flips the order, i.e. shows the image 2 reprojected into image 1.\n\n\n- `Save points` -- saves (overwrites) the correspondences into the text file.\n\n- `Prev` -- Loads and shows previous image pair to label. If the `CorrespondenceAnnotator` was initialized with `save_on_next=True`, the current pair correspondences are saved before the switch. Change is not cyclical, so the button does nothing on 1st image pair\n\n- `Next` -- Loads and shows next image pair to label. If the `CorrespondenceAnnotator` was initialized with `save_on_next=True`, the current pair correspondences are saved before the switch.Change is not cyclical, so the button does nothing on last image pair.\n\n- `CLAHE` -- Images are shown with enhanced contrast with [CLAHE](http://amroamroamro.github.io/mexopencv/opencv/clahe_demo_gui.html) algorithm.\n\n\n\n- `Pick 4 points for homography` -- Special mode. User picks 4 points in one image, which define new fronto-parallel view. This mode helps for labeling obscure views, see example below. The order of the points: top-left -> top-right -> bottom right -> bottom left. After the 4 point picked, the mode is switched off, so user needs to click the button again if she wants to rectify another image.\n\n- `ResetView` -- Resets any recifications or zoom done to images.\n\n\n\n### Rectification picking mode example\n\n3 points are selected\n\n\nAll points are selected and image1 is rectified to the rectangle, defined by the selected points\n\n",
"bugtrack_url": null,
"license": "Apache Software License 2.0",
"summary": "Matplotlib-based tool for labeling the two-view correspondences",
"version": "0.1.5",
"project_urls": {
"Homepage": "https://github.com/ducha-aiki/pixelstitch/tree/master/"
},
"split_keywords": [
"local features",
" correspondences",
" ransac",
" image matching",
" wbs",
" wxbs"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "1a83bf7923aaf791b6a4e671f71570fa56d50be4c0ef2dc8aced1685462e0531",
"md5": "5272de6b0cb6595a7213e60d5ff1587c",
"sha256": "2e1e9d9248fcb624de46547283625f0a71035827dd7518b64ea66f09154d9e1b"
},
"downloads": -1,
"filename": "pixelstitch-0.1.5-py3-none-any.whl",
"has_sig": false,
"md5_digest": "5272de6b0cb6595a7213e60d5ff1587c",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.8",
"size": 20358,
"upload_time": "2024-08-06T07:05:58",
"upload_time_iso_8601": "2024-08-06T07:05:58.483436Z",
"url": "https://files.pythonhosted.org/packages/1a/83/bf7923aaf791b6a4e671f71570fa56d50be4c0ef2dc8aced1685462e0531/pixelstitch-0.1.5-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "d0621e40420b1eeb4f2c4c8cbecd4d41ddffdfbac7275c56a46c91bc6aef1962",
"md5": "dc37b6d20b491f847bce22e5161891fd",
"sha256": "7d5c9596153b1e05ec36ed0433b1203137ac954b1cab5d6c3ac9c6829c62f166"
},
"downloads": -1,
"filename": "pixelstitch-0.1.5.tar.gz",
"has_sig": false,
"md5_digest": "dc37b6d20b491f847bce22e5161891fd",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.8",
"size": 22259,
"upload_time": "2024-08-06T07:06:00",
"upload_time_iso_8601": "2024-08-06T07:06:00.249030Z",
"url": "https://files.pythonhosted.org/packages/d0/62/1e40420b1eeb4f2c4c8cbecd4d41ddffdfbac7275c56a46c91bc6aef1962/pixelstitch-0.1.5.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-08-06 07:06:00",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "ducha-aiki",
"github_project": "pixelstitch",
"travis_ci": false,
"coveralls": false,
"github_actions": true,
"lcname": "pixelstitch"
}
JSON
