PySceneDetect v0.5 New CLI & API Feedback

Hello all;

Please free to leave any feedback or suggestions you might have regarding the latest release of PySceneDetect (v0.5). This release includes a significantly refactored Python API as well as a completely redone command-line interface (CLI).

Feedback for both the new API and CLI is requested here.

Thank you all again for your help and support with respect to preparing this release - really appreciate everyone's feedback!

@Breakthrough

Hi @Breakthrough,

I have been doing some work recently with the 0.5 API and overall, I think the new structure is really well thought out. There is just one minor thing I want to point out. Would it be possible to bring the StatsManager class into the base import scenedetect call? As it is right now, this is what getting ready to analyze a video looks like:

import scenedetect

video_mgr = scenedetect.VideoManager([video_file])
stats_mgr = scenedetect.stats_manager.StatsManager()
scene_mgr = scenedetect.SceneManager(stats_mgr)

This would be fixed by adding an import statement to scenedetect.__init__.py to import StatsManager along with the VideoManager and SceneManager classes.

This would be a low priority for sure (I said it was minor). Thanks for your hard work!

@wjs018

Hi @wjs018;

Yes absolutely, good catch. I intended for it to be there, I must have missed adding it.

Thanks for the report, I will definitely include a fix for this in the next release. Likewise, thanks for your interest in the project!

Note that going forwards, the capabilities of DVR-Scan will be absorbed into PySceneDetect, with the appropriate API changes being made in v0.6 going forwards. There may be some changes to the scene detection algorithm interfaces, but I don't think any of the changes will break existing code, just add new functionality.

That being said, thank you for your input - I've pinned this issue to gain some last minute visibility as I plan what's going to be included in the v0.6 release (the biggest being support of SceneStats).

Random, but API changes to the detectors will be required to implement camera flash suppression. Specifically, a lookahead buffer will be required (of size N frames). My idea is to allow detectors to specify, during construction, the amount of lookahead they require, and will let the SceneManager know. That way, the SceneManager can pre-buffer that many frames and provide them to each detector, rather than one frame at a time (and requiring individual detectors to buffer frames manually, as is done currently).

I would like to apply this in both directions, i.e. allow detectors to search +/- N frames in time from the current frame. This will dramatically increase memory and CPU utilization, so N must be a user tunable parameter, as certain source material will benefit greatly from it, at the expense of processing speed.

I keep getting an error, just for save-images

scenedetect -i '/home/santiago/Documents/NewConsultoriaMarketing/filename.mkv' detect-content save-images

[PySceneDetect] Loaded 1 video, framerate: 29.97 FPS, resolution: 1280 x 720
[PySceneDetect] Downscale factor set to 5, effective resolution: 256 x 144
[PySceneDetect] Image output format set: JPEG
[PySceneDetect] Detecting scenes...
100%|████████████████████████████████| 9680/9680 [00:24<00:00, 388.39frames/s]
[PySceneDetect] Processed 9680 frames in 24.9 seconds (average 388.35 FPS).
[PySceneDetect] Detected 9 scenes, average shot length 35.9 seconds.
[PySceneDetect] Comma-separated timecode list:
00:00:00.734,00:00:01.802,00:03:33.146,00:03:38.652,00:04:21.094,00:05:18.752,00:05:19.886,00:05:20.954
[PySceneDetect] Downscale factor set to 1, effective resolution: 1280 x 720
[PySceneDetect] Generating output images (3 per scene)...
0%| | 0/27 [00:00<?, ?images/s]Traceback (most recent call last):
File "/home/santiago/.virtualenvs/videotracking/bin/scenedetect", line 8, in
sys.exit(main())
File "/home/santiago/.virtualenvs/videotracking/local/lib/python2.7/site-packages/scenedetect/main.py", line 58, in main
cli.main(obj=cli_ctx) # Parse CLI arguments with registered callbacks.
File "/home/santiago/.virtualenvs/videotracking/local/lib/python2.7/site-packages/click/core.py", line 792, in main
ctx.exit()
File "/home/santiago/.virtualenvs/videotracking/local/lib/python2.7/site-packages/click/core.py", line 404, in exit
self.close()
File "/home/santiago/.virtualenvs/videotracking/local/lib/python2.7/site-packages/click/core.py", line 494, in close
cb()
File "/home/santiago/.virtualenvs/videotracking/local/lib/python2.7/site-packages/scenedetect/cli/context.py", line 397, in process_input
output_dir=self.image_directory)
File "/home/santiago/.virtualenvs/videotracking/local/lib/python2.7/site-packages/scenedetect/cli/context.py", line 225, in _generate_images
for start, end in scene_list
AttributeError: 'list' object has no attribute 'stop'
0%| | 0/27 [00:00<?, ?images/s]

@santiagodemierre

Hi @santiagodemierre;

Good find! It looks like this only affects Python 2.7. Are you able to upgrade to Python 3?

I have filed a new issue (#174) to track the resolution for that. This is definitely a bug - I'll try to roll a fix out promptly (may have to roll out v0.5.4).

Thank you for letting me know, and sorry about the inconvenience!

Thank you all for your feedback and support. All future CLI and API changes will now be targeted for PySceneDetect v0.6. Any additional releases under v0.5.x will strictly be to address any major bugs.

Users wishing to provide feedback on upcoming planned API and CLI changes are free to do so in #177. I especially encourage users using the Python API to review the upcoming changes, and to provide any feedback if a transition plan is required to implement these changes in a non-breaking way. Although most of the planned changes will not break the most widely used APIs, your feedback is important to ensuring that the new changes take into account most potential use cases, support future extensibility, and can be integrated in a way that has minimal impact on existing dependencies.

Any additional bug reports found in existing v0.5 versions should be submitted as a new issue.

Navigation Menu

Repository navigation

PySceneDetect v0.5 New CLI & API Feedback #72

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

PySceneDetect v0.5 New CLI & API Feedback #72

Description

Activity

wjs018 commented on Nov 24, 2018

Breakthrough commented on Jan 7, 2019

Breakthrough commented on May 20, 2019

Breakthrough commented on Jun 27, 2020

santiagodemierre commented on Aug 6, 2020

Breakthrough commented on Aug 8, 2020

Breakthrough commented on Aug 8, 2020

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions