[WIP] short blogpost on tooling around processing video datasets #2631
+151
−0
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change | ||||
|---|---|---|---|---|---|---|
| @@ -0,0 +1,151 @@ | ||||||
| --- | ||||||
| title: "Build awesome datasets for video generation" | ||||||
| thumbnail: /blog/assets/vid_ds_scripts/thumbnail.png | ||||||
| authors: | ||||||
| - user: hlky | ||||||
| - user: sayakpaul | ||||||
| --- | ||||||
|
|
||||||
| # Build awesome datasets for video generation | ||||||
|
|
||||||
| Tooling for image generation datasets is well established, with [`img2dataset`](https://github.com/rom1504/img2dataset) covering large scale and various community guides, scripts and UIs covering the small scale. | ||||||
|
There was a problem hiding this comment.
Suggested change
|
||||||
|
|
||||||
| Our goal is to make tooling for video generation datasets as established by creating open video dataset scripts suited for small scale, with [`video2dataset`](https://github.com/iejMac/video2dataset) covering large scale. | ||||||
|
There was a problem hiding this comment.
Suggested change
I'm not sure if this is what you want to say, or rather than we'll focus on the small scale because the large scale is "solved". |
||||||
|
|
||||||
| *“If I have seen further it is by standing on the shoulders of giants”* | ||||||
|
|
||||||
| In this post, we will provide a overview of the tooling we are developing to make it easy for the community to build their own datasets for fine-tuning video generation models. If you cannot wait to get started already, we welcome you to check out the codebase [here](https://github.com/huggingface/video-dataset-scripts/tree/main/video_processing). | ||||||
|
There was a problem hiding this comment.
Suggested change
|
||||||
|
|
||||||
| **Table of contents** | ||||||
|
|
||||||
| TODO | ||||||
|
|
||||||
| ## Tooling | ||||||
|
|
||||||
| Typically, video generation is conditioned on natural language text prompts such as: "A cat walks on the grass, realistic style". Then in a video, there are a number of qualitative aspects for controllability and filtering, like so: | ||||||
|
|
||||||
| * Motion | ||||||
| * Aesthetics | ||||||
| * Presence of watermarks | ||||||
| * Presence of NSFW content | ||||||
|
|
||||||
| Video generation models are only as good as the data they are trained on. Therefore, these aspects become crucial when curating the datasets for training/fine-tuning. | ||||||
|
|
||||||
| Our 3 stage pipeline draws inspiration from works like [Stable Video Diffusion](https://arxiv.org/abs/2311.15127), [LTX-Video](https://arxiv.org/abs/2501.00103), and their data pipelines. | ||||||
|
|
||||||
| ### Stage 1 (Acquisition) | ||||||
|
|
||||||
| Like [`video2dataset`](https://github.com/iejMac/video2dataset) we opt to use [`yt-dlp`](https://github.com/yt-dlp/yt-dlp) for downloading videos. | ||||||
|
|
||||||
| We create a script `Video to Scenes` to split long videos into short clips. | ||||||
|
|
||||||
| ### Stage 2 (Pre-processing/filtering) | ||||||
|
|
||||||
| #### Extracted frames | ||||||
|
|
||||||
| - detect watermarks with [LAION-5B-WatermarkDetection](https://github.com/LAION-AI/LAION-5B-WatermarkDetection) | ||||||
| - predict an aesthetic score with [improved-aesthetic-predictor](https://github.com/christophschuhmann/improved-aesthetic-predictor) | ||||||
| - detect presence of NSFW content with [Falconsai/nsfw_image_detection](https://huggingface.co/Falconsai/nsfw_image_detection) | ||||||
|
|
||||||
| #### Entire video | ||||||
|
|
||||||
| - predict a motion score with [OpenCV](https://docs.opencv.org/3.4/d4/dee/tutorial_optical_flow.html) | ||||||
|
There was a problem hiding this comment. I think we should state what it is and what it will be used for. The link to optical flow methods might be a bit confusing in my opinion. |
||||||
|
|
||||||
| ### Stage 3 (Processing) | ||||||
|
|
||||||
| Florence-2 [`microsoft/Florence-2-large`](http://hf.co/microsoft/Florence-2-large) to run `<CAPTION>`, `<DETAILED_CAPTION>`, `<DENSE_REGION_CAPTION>` and `<OCR_WITH_REGION>`. | ||||||
|
|
||||||
| We can bring in any other captioner in this regard. We can also caption the entire video (e.g., with a model like [Qwen2.5](https://huggingface.co/docs/transformers/main/en/model_doc/qwen2_5_vl)) as opposed to captioning individual frames. | ||||||
|
Comment on lines
+56
to
+58
There was a problem hiding this comment. Are we captioning individual frames with Florence 2? What do those tags In general, I think we should previously introduce the task earlier in the post: we are going to grab a lot of videos from the Internet, and we are going to caption them so that the models to be trained have enough information to match text descriptions with frames/sequences of frames/full videos. And of course we need to filter bad quality content, videos that are too short or too long, etc. |
||||||
|
|
||||||
| ## Filtering examples | ||||||
|
|
||||||
| In the [dataset](https://huggingface.co/datasets/finetrainers/crush-smol) for the model [`finetrainers/crush-smol-v0`](https://hf.co/finetrainers/crush-smol-v0) we filtered on `pwatermark < 0.1` and `aesthetic > 5.5`. This highly restrictive filtering resulted in 47 videos out of 1493 total. | ||||||
|
There was a problem hiding this comment. This one appears to have been captioned with Qwen actually, but we said we'd use Florence. I'd explain that this is an example dataset we obtained using this process. |
||||||
|
|
||||||
| Let's review the example frames from `pwatermark` - two with text have scores of 0.69 and 0.61, the "toy car with a bunch of mice in it" scores 0.60 then 0.17 as the toy car is crushed. All example frames were filtered by `pwatermark < 0.1`. `pwatermark` is effective at detecting text/watermarks however the score gives no indication whether it is a text overlay or a toy car's license plate. Our filtering required all scores to be below the threshold, an average across frames would be a more effective strategy for `pwatermark` with a threshold of around 0.2 - 0.3. | ||||||
|
|
||||||
| Let's review the example frames from aesthetic scores - the pink castle initially scores 5.5 then 4.44 as it is crushed, the action figure scores lower at 4.99 dropping to 4.84 as it is crushed and the shard of glass scores low at 4.04. Again fitlering required all scores to be below the threshold, in this case using the aesthetic score from the first frame only would be a more effective strategy. | ||||||
|
Comment on lines
+64
to
+66
There was a problem hiding this comment. I'm a bit lost here. Screenshots would go a long way imo. |
||||||
|
|
||||||
| If we review [`finetrainers/crush-smol`](https://huggingface.co/datasets/finetrainers/crush-smol) we can notice that many of the objects being crushed are round or rectangular and colorful which is similar to our findings in the example frames. Aesthetic scores can be useful yet have a bias that will potentially filter out good data when used with extreme thresholds like > 5.5. It may be more effective as a filter for bad content than good with a minimum threshold of around 4.25 - 4.5. | ||||||
|
|
||||||
| We provide some visual examples for each filters as well as the captions from Florence-2. | ||||||
|
|
||||||
| ### Watermarks | ||||||
|
There was a problem hiding this comment. Ah, the example images are here! I'd put them above and maybe interwoven with the text explanation. |
||||||
|
|
||||||
| | **`pwatermark`** | **Image** | | ||||||
| |:----------:|:-----:| | ||||||
| | 0.69 |  | | ||||||
| | 0.61 |  | | ||||||
| | 0.60 |  | | ||||||
| | 0.17 |  | | ||||||
|
|
||||||
|
|
||||||
| ### Aesthetics | ||||||
|
|
||||||
| | **Aesthetic** | **Image** | | ||||||
| |:---------:|:-----:| | ||||||
| | 5.50 |  | | ||||||
| | 4.44 |  | | ||||||
| | 4.99 |  | | ||||||
| | 4.87 |  | | ||||||
| | 4.84 |  | | ||||||
| | 4.04 |  | | ||||||
|
|
||||||
| ### OCR/Caption | ||||||
|
|
||||||
| <table> | ||||||
| <tr> | ||||||
| <th>Image</th> | ||||||
| <th>Caption</th> | ||||||
| <th>Detailed Caption</th> | ||||||
| </tr> | ||||||
| <tr> | ||||||
| <td> | ||||||
| <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/vid_ds_scripts/-IvRtqwaetM-Scene-003_0%201.jpg" | ||||||
| alt="Toy Car with Mice"> | ||||||
| </td> | ||||||
| <td>A toy car with a bunch of mice in it.</td> | ||||||
| <td>The image shows a blue toy car with three white mice sitting in the back of it, driving down a road with a green wall in the background.</td> | ||||||
| </tr> | ||||||
| </table> | ||||||
|
|
||||||
| <table> | ||||||
| <tr> | ||||||
| <th>With OCR labels</th> | ||||||
| <th>With OCR and region labels</th> | ||||||
| </tr> | ||||||
| <tr> | ||||||
| <td> | ||||||
| <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/vid_ds_scripts/image_with_ocr_labels.jpg" | ||||||
| alt="OCR labels"> | ||||||
| </td> | ||||||
| <td> | ||||||
| <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/vid_ds_scripts/image_with_ocr_and_region_labels.jpg" | ||||||
| alt="OCR and region labels"> | ||||||
| </td> | ||||||
| </tr> | ||||||
| </table> | ||||||
|
|
||||||
|
|
||||||
| ## Putting this tooling to use 👨🍳 | ||||||
|
|
||||||
| We have created various datasets with the tooling in an attempt to generate cool video effects, similar to the [Pika Effects](https://pikartai.com/effects/): | ||||||
|
|
||||||
| - [Cakeify](https://huggingface.co/datasets/finetrainers/cakeify-smol) | ||||||
| - [Crush](https://huggingface.co/datasets/finetrainers/crush-smol) | ||||||
|
|
||||||
| We then used these datasets to fine-tune the [CogVideoX-5B](https://huggingface.co/THUDM/CogVideoX-5b) model using [`finetrainers`](https://github.com/a-r-r-o-w/finetrainers). Below is an example output from [`finetrainers/crush-smol-v0`](https://huggingface.co/finetrainers/crush-smol-v0): | ||||||
|
|
||||||
|
|
||||||
| <figure class="image flex flex-col items-center text-center m-0 w-full"> | ||||||
| <video | ||||||
| alt="demo4.mp4" | ||||||
| autoplay loop autobuffer muted playsinline | ||||||
| > | ||||||
| <source src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/vid_ds_scripts/assets_output_0.mp4" type="video/mp4"> | ||||||
| </video> | ||||||
| <figcaption>Prompt: <i>DIFF_crush A red candle is placed on a metal platform, and a large metal cylinder descends from above, flattening the candle as if it were under a hydraulic press. The candle is crushed into a flat, round shape, leaving a pile of debris around it.</i></figcaption> | ||||||
| </figure> | ||||||
|
|
||||||
| ## Call to action | ||||||
|
There was a problem hiding this comment. I wouldn't use "call to action", as I think it's a marketing term. Maybe "Your Turn", or "Follow Along", or "Conclusion" could work. |
||||||
|
|
||||||
| We hope this tooling gives you a headstart to create small and high-quality video datasets for your own custom applications. We will continue to add more useful filters to the repository, so, please keep an eye out. Your contributions are also more than welcome 🤗 | ||||||
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
TODO: change.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We also need the changes to
_blog.ymletc.