I have just open-sourced a literature summary on ML-based content moderation and multimodal content rating! It includes various sources for audio, text, video modalities, links to datasets and related Python tools.
Github: github.com/fcakyon/conten
Arxiv: arxiv.org/abs/2212.04533
Follow
Click to Follow fcakyon
fatih
@fcakyon
ml engineer by day, phd student by night
interested in opensource & multi-modal learning
github.com/fcakyon
linkedin.com/in/fcakyonJoined February 2020
fatih’s Tweets
"Imagine learning a textbook with no figures."
Multimodal chain-of-thought (Multimodal-CoT) in Language Models
- Outperform GPT-3.5 by 16% (75%->91%) and surpass human performance on ScienceQA
- Less than 1B params (so you can train more easily)
- Code & model released
[1/6]
6
110
444
Show this thread
Google announces Dreamix: a model that generates videos when given:
- video + prompt (Video editing)
- input images + prompt (Subject Driven Generation)
- input image + prompt (Image-toVideo
0:48
404.2K views
90
1,848
8,114
Check our latest opensource Python package. It contains many utilities for medical signal processing 💯
Medikal sinyal isleme icin bircok hazir fonksiyon iceren kutuphanemiz yayinda
github: github.com/obss/BIOBSS
docs: biobss.readthedocs.io/en/latest/
Quote Tweet
We have open sourced the #biomedical #signalprocessing library we have been using as a team with slight improvements. Recommended especially if you are working with ECG, PPG, EDA or ACC signals. Comparison with other libraries in the pic #opensource github.com/obss/BIOBSS (1)
Show this thread
4
6
Thanks for highlighting my zero-shot and supervised video classification apps!
Quote Tweet
At @huggingface, we've been making efforts to integrate SOTA vision and multimodal models.
Check out this blog post by @RisingSayak and @alaradirik for a nice overview 
huggingface.co/blog/cv_state
huggingface.co/blog/cv_state1
3
A blog post I wrote about how to approach drone detection and possibly other #computervision taks. It taks about using sa cascaded system and using synthetic data properly. #ai #drone medium.com/codable/challe
Would love to her what you think.
1
6
Great source on finetuning huggingface object detection models on huggingface object detections datasets:
4
A team from just released “balance”, an open source Python package for adjusting biased samples.
It's a simple easy-to-use framework for weighting data and evaluating its biases with and without adjustments.
Code: github.com/facebookresear
Blog: import-balance.org/blog/2023/01/0
9
260
997
New useful YOLOv5 models and object detection datasets are available on the hub. Check them if you are dealing with game analytics 👍🏻
Oyun analitiği veya otomasyonuyla ilgilenenler bu model ve verisetlerini çalışmalarında kullanabilir 👌🏻
Quote Tweet
Selamlar herkese,
Eğittiğim bazı YOLOv5 modellerini ve verisetlerini @huggingface üzerinde paylaştım.
Valorant Model: huggingface.co/keremberke/yol
CSGO Model: huggingface.co/keremberke/yol
Clash of Clans Model: huggingface.co/keremberke/yol
#valorant #csgo #clashofclans #ml
Show this thread
7
Icerik moderasyonu alaninda bircok veriseti, makale ve aracin linklerini ve ozetlerini derledigim literatur taramasini acik kaynak olarak paylastim.
Eklenmesini istediginiz calismalar varsa yazabilirsiniz 👍
Github: github.com/fcakyon/conten
Arxiv: arxiv.org/abs/2212.04533
GIF
1
3
Show this thread
There are many large, interesting datasets across different sectors - e.g., medicine, law, finance. Rather than relying on a single 100B+ parameter foundation model, we think there’s a lot of value that can be captured by <10B parameter models trained on domain-specific datasets.
1
12
77
Show this thread
📣 CRFM announces PubMedGPT, a new 2.7B language model that achieves a new SOTA on the US medical licensing exam. The recipe is simple: a standard Transformer trained from scratch on PubMed (from The Pile) using on the MosaicML Cloud, then fine-tuned for the QA task.
40
391
1,556
Show this thread
I am glad you find it useful!
Quote Tweet
@fcakyon thanks for creating yolov5-pip! It addresses two pain points I ran into when using yolov5 - being able to pip install and access as a python api.
1
We just open sourced our tokenizer! This should help all users of our API better understand their usage. It's also incredibly fast – 3-6x faster than other open source alternatives!
49
316
2,497
hub'daki video siniflandirma modellerini deneyebileceginiz bir demo olusturdum.
transformers kutuphanesine yeni eklenen video-classification pipeline'ini nasil kullanacaginizi merak ediyorsaniz kodu inceleyebilirsiniz 👍
Demo: huggingface.co/spaces/fcakyon
3
Show this thread
I have just published a demo that lets you try many video classification models from the hub 📽
It utilizes the newly introduced video-classification pipeline that developed!
Demo: huggingface.co/spaces/fcakyon
GIF
1
10
40
Show this thread
Bonus: You can use the video-transformers package if you want to fine-tune pretrained TimeSformer models on your custom dataset: github.com/fcakyon/video-
1
1
2
Show this thread
Happy to announce that I have contributed the TimeSformer architecture to the transformers library and it is available in the main branch!
Thank you for the guidance 👍
demo: huggingface.co/spaces/fcakyon
docs: huggingface.co/docs/transform
GIF
2
24
91
Show this thread
Buyuk goruntulerdeki kucuk objeleri tespit etmek artik daha da kolay!
Tum YOLO modelleri icin egitim/tespit ve deployment destegi sunan MMYOLO kutuphanesinin son versiyonuyla birlikte SAHI entegrasyonu geldi🚀
Bu linkten deneyebilirsiniz: github.com/open-mmlab/mmy
2
1
Show this thread
It is now easier to detect small objects on large images (as satellite imagery)!
MMYOLO v0.2.0 (that supports training and inference for all YOLO models) adds SAHI integration for large image inference 🔥
Try it at this url: github.com/open-mmlab/mmy
2
12
56
Show this thread
Geliştirdiğimiz sistem her yerde haber olmuş 😎
Quote Tweet
1
1
4
We are excited to announce the release of Stable Diffusion Version 2!
Stable Diffusion V1 changed the nature of open source AI & spawned hundreds of other innovations all over the world. We hope V2 also provides many new possibilities!
Link → stability.ai/blog/stable-di
165
2,402
8,451
Show this thread
It seems there is no support in mmengine which is the core package of mmdet 3.x mmyolo mmrotate 1.x etc.
Do you have any plans to open a pr to mmengine adding support?
Here is the related thread:
3
1
2
Zero shot object detection is modern day magic
I'm using it to expand my dataset by generating free annotated images
Here's my workflow
Let’s dive in:🧵
23
96
529
Show this thread
Automated
Congrats to openai/whisper, which has grown by 804 stars in the last 7 days and has reached 14.9K stars.
Thanks to the contributors:
ossinsight.io/analyze/openai
#python
3
2
![Markdown file with the following contents:
```mermaid
stateDiagram-v2
[*] --> Still
Still --> [*]
Still --> Moving
Moving --> Still
Moving --> Crash
Crash --> [*]
```](https://pbs.twimg.com/media/FhD3G6iXoAErMGo?format=png&name=small)
📣 New 'Transfer Learning' Flax guide!
We added a new guide that shows how to use models from 's transformers library, perform parameter surgery, and freeze parameters with optax or perform fine-tunning with different learning rates.
flax.readthedocs.io/en/latest/guid
2
27
137
Show this thread
Excited to share that my day-long workshop (a short course) on #ExplainableAI is now publicly available as a five-part youtube video lecture series.
Link to video lectures: lnkd.in/gzfmJug9
Link to slides: lnkd.in/e_RsBVPx
#AI #ML
23
459
2,041
Show this thread
Our test satellite İstanbul is getting ready for its mission. Next is environmental test. Share the excitement of Team Istanbul of Hello Space.🚀🚀🚀
#HelloSpace
5
33
240
When a SAHI bug is fixed but you dont know how
Quote Tweet
1
😍 Lama Cleaner: A free and open-source inpainting tool powered by SOTA AI model
🎁 Completely Free and Open-Source
• Fully self-hosted
• Multiple SOTA AI models
• Classical image inpainting algorithm
• CPU & GPU
• Various inpainting strategy
• Run as a Desktop APP
0:14
107.7K views
38
573
3,466
Show this thread
That feeling 😅
Quote Tweet
So when trying to memify this lesson you bet I was listening hard
Show this thread
1
1
1
Step 5: Enter Slicing Aided Hyper Inference (SAHI) technique
Using high resolution images I was able to detect almost all the bounding boxes.
1
4
6
Show this thread
This sweater developed by the University of Maryland is an invisibility cloak against AI.
It uses "adversarial patterns" to stop AI from recognizing the person wearing it.
1:11
1.1M views
587
6,656
27K
Show this thread
A usual twitter drama
4
8
114