| Name | perfectcrop JSON |
| Version |
1.1.1
JSON |
| download |
| home_page | |
| Summary | A video cropping tool for artists that detects an object, finds the average center pixel of the bounding box, and crops out from center without a decrease in video quality. |
| upload_time | 2024-03-04 05:44:41 |
| maintainer | |
| docs_url | None |
| author | Rebecca Picanso |
| requires_python | >=3.8,<4.0 |
| license | |
| keywords |
object detection
video
|
| VCS |
|
| bugtrack_url |
|
| requirements |
No requirements were recorded.
|
| Travis-CI |
No Travis.
|
| coveralls test coverage |
No coveralls.
|
# Perfect Crop
Perfect Crop will locate a detected object within a video clip, find its average location, and create a cropped clip from the center. Object detection models splice videos into individual frames & then run detection on static images - Perfect Crop transmutes this into workable video.
This tool is most useful when used on large clip sets (200+). It was built for the quick processing of over 50 hours of live-stream footage!
As a test example, searching the video for "cell phone":
Source Clip | Detected Object
:-------------------------:|:-------------------------:
| 
Perfect Crop's algorithm was built around Hugging Face's [hustvl/yolo](https://www.google.com/search?q=hustvl%2Fyolo-tiny&rlz=1C5CHFA_enUS997US998&oq=yolostiny+hu&gs_lcrp=EgZjaHJvbWUqCggBEAAYChgWGB4yBggAEEUYOTIKCAEQABgKGBYYHjINCAIQABiGAxiABBiKBdIBCDMwNTRqMGo0qAIAsAIA&sourceid=chrome&ie=UTF-8) sets, a series of vision transformer models trained on the [COCO](https://cocodataset.org/#home) dataset. It's computationally light on any system. At the moment, the algorithm is set only for models using YOLO bounding box annotations - so if you want to swap out the model, keep that in mind.
Calculating the center pixel of a bounding box requires backward engineering of detection algorithms - I'm excited to release support for other annotations in v2.
---
Perfect Crop can also label scenes with content and object locations, as well as index a video & splice it into individual scenes without quality loss.
```
perfectcrop accepts the following inputs:
-p [video or directory] | crop a video or directory of clips to select label (runs perfect_crop.py)
-s [video] | create .json file identifying shots (scene changes) in a long form video. shots.py will then splice the video into individual clips (or scenes).
-vs [video or directory] | standardizes video length, codecs, etc. with ffmpeg bash. needed for concatenation!
-g [video or directory] | creates a dynamic, randomized moving video grid with ffmpeg bash commands. this is a legacy file from Court Laureate, not necessary for the project itself :)
```
Due to the sensitive nature of working with video, Perfect Crop does not currently support multiple command line options.
Please note that perfectcrop -s needs to be run in the same directory as the source video for shot splicing to occur.
potential errors:
```
* global loadsave -> thankfully, this is harmless and most often caused by minute differences in codecs or issues with local reading. I'll hide it.
* video not found -> please check permissions, you can do this by running ls -al on the directory & open with chmod commands.
* file limit hit -> grid specific error. please run 'ulimit -n' in terminal to see your current limit, then ulimit -n [whatever] to increase the size to whatever you see fit. please note that this is isolated to your terminal window.
```
---
**Made for Court Laureate, a short film screened at LARPA, December 2023, NY NY.
Although one cannot patent an algorithm, please respect the open-source nature of this project & all derivatives.**
Raw data
{
"_id": null,
"home_page": "",
"name": "perfectcrop",
"maintainer": "",
"docs_url": null,
"requires_python": ">=3.8,<4.0",
"maintainer_email": "",
"keywords": "object detection,video",
"author": "Rebecca Picanso",
"author_email": "rebeccapicansot@gmail.com",
"download_url": "https://files.pythonhosted.org/packages/a1/59/c3e2fd6deb1bfc4cd2e0a7dd79cb2a2c211cf70e281a28c893cf91b86b32/perfectcrop-1.1.1.tar.gz",
"platform": null,
"description": "# Perfect Crop\n\nPerfect Crop will locate a detected object within a video clip, find its average location, and create a cropped clip from the center. Object detection models splice videos into individual frames & then run detection on static images - Perfect Crop transmutes this into workable video.\n\nThis tool is most useful when used on large clip sets (200+). It was built for the quick processing of over 50 hours of live-stream footage!\n\nAs a test example, searching the video for \"cell phone\":\nSource Clip | Detected Object\n:-------------------------:|:-------------------------:\n| \n\nPerfect Crop's algorithm was built around Hugging Face's [hustvl/yolo](https://www.google.com/search?q=hustvl%2Fyolo-tiny&rlz=1C5CHFA_enUS997US998&oq=yolostiny+hu&gs_lcrp=EgZjaHJvbWUqCggBEAAYChgWGB4yBggAEEUYOTIKCAEQABgKGBYYHjINCAIQABiGAxiABBiKBdIBCDMwNTRqMGo0qAIAsAIA&sourceid=chrome&ie=UTF-8) sets, a series of vision transformer models trained on the [COCO](https://cocodataset.org/#home) dataset. It's computationally light on any system. At the moment, the algorithm is set only for models using YOLO bounding box annotations - so if you want to swap out the model, keep that in mind.\n\nCalculating the center pixel of a bounding box requires backward engineering of detection algorithms - I'm excited to release support for other annotations in v2.\n\n---\nPerfect Crop can also label scenes with content and object locations, as well as index a video & splice it into individual scenes without quality loss.\n\n```\nperfectcrop accepts the following inputs:\n\n -p [video or directory] | crop a video or directory of clips to select label (runs perfect_crop.py)\n\n -s [video] | create .json file identifying shots (scene changes) in a long form video. shots.py will then splice the video into individual clips (or scenes).\n \n -vs [video or directory] | standardizes video length, codecs, etc. with ffmpeg bash. needed for concatenation!\n \n -g [video or directory] | creates a dynamic, randomized moving video grid with ffmpeg bash commands. this is a legacy file from Court Laureate, not necessary for the project itself :)\n```\n\nDue to the sensitive nature of working with video, Perfect Crop does not currently support multiple command line options.\nPlease note that perfectcrop -s needs to be run in the same directory as the source video for shot splicing to occur.\n\npotential errors:\n```\n * global loadsave -> thankfully, this is harmless and most often caused by minute differences in codecs or issues with local reading. I'll hide it.\n * video not found -> please check permissions, you can do this by running ls -al on the directory & open with chmod commands.\n * file limit hit -> grid specific error. please run 'ulimit -n' in terminal to see your current limit, then ulimit -n [whatever] to increase the size to whatever you see fit. please note that this is isolated to your terminal window.\n```\n\n---\n**Made for Court Laureate, a short film screened at LARPA, December 2023, NY NY.\nAlthough one cannot patent an algorithm, please respect the open-source nature of this project & all derivatives.**\n\n\n",
"bugtrack_url": null,
"license": "",
"summary": "A video cropping tool for artists that detects an object, finds the average center pixel of the bounding box, and crops out from center without a decrease in video quality.",
"version": "1.1.1",
"project_urls": null,
"split_keywords": [
"object detection",
"video"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "255a3c984bc77c84bc26c167e0811ac9a141ed4bc5be14fb41bcff99d2ab1d6c",
"md5": "6491323499c7b8f81c3c6c72eb35b898",
"sha256": "bf2b004e4c9005e8e9953e965b746744784b3129ae106fe1af0f528b4629bdcd"
},
"downloads": -1,
"filename": "perfectcrop-1.1.1-py3-none-any.whl",
"has_sig": false,
"md5_digest": "6491323499c7b8f81c3c6c72eb35b898",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.8,<4.0",
"size": 21888,
"upload_time": "2024-03-04T05:44:39",
"upload_time_iso_8601": "2024-03-04T05:44:39.300306Z",
"url": "https://files.pythonhosted.org/packages/25/5a/3c984bc77c84bc26c167e0811ac9a141ed4bc5be14fb41bcff99d2ab1d6c/perfectcrop-1.1.1-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "a159c3e2fd6deb1bfc4cd2e0a7dd79cb2a2c211cf70e281a28c893cf91b86b32",
"md5": "5427ebf88a505b5917477716dbec9326",
"sha256": "3946be17babde08087ff7f27d25e81188a07f50b4d0edb7da6ecf9ba90c7b0a6"
},
"downloads": -1,
"filename": "perfectcrop-1.1.1.tar.gz",
"has_sig": false,
"md5_digest": "5427ebf88a505b5917477716dbec9326",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.8,<4.0",
"size": 21274,
"upload_time": "2024-03-04T05:44:41",
"upload_time_iso_8601": "2024-03-04T05:44:41.129100Z",
"url": "https://files.pythonhosted.org/packages/a1/59/c3e2fd6deb1bfc4cd2e0a7dd79cb2a2c211cf70e281a28c893cf91b86b32/perfectcrop-1.1.1.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-03-04 05:44:41",
"github": false,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"lcname": "perfectcrop"
}
