The integration supports low-VRAM fine-tuning on constrained hardware.
RF-DETR just landed to @huggingface transformers 🥵🔥
sota real-time detection & segmentation models by @roboflow 💜
> play with our real-time demo > fine-tune the models on your use case with our tutorials (takes a toaster's VRAM) > or just hand them to your agents 😄
Many users expressed excitement about Roboflow's RF-DETR model now being available in Hugging Face Transformers because of its impressive real-time object detection quality and potential uses.
No Digg Deeper questions have been answered for this story yet.
#861
RF-DETR is now available in @huggingface transformers
state of the art in both detection and segmentation, outperforming YOLO architectures
- checkpoints: https://huggingface.co/Roboflow/models - demo: https://huggingface.co/spaces/huggingface-projects/rf-detr-realtime-webcam - docs: https://huggingface.co/collections/merve/rfdetr
to celebrate it we ship > real-time webcam demo > fine-tuning notebook on satellite imagery segmentation > another notebook on phone UI detection
tutorials → https://github.com/merveenoyan/smol-vision/tree/main/rf-detr models and the demo → https://huggingface.co/collections/merve/rfdetr docs → https://huggingface.co/docs/transformers/main/model_doc/rf_detr
RF-DETR just landed to @huggingface transformers 🥵🔥
sota real-time detection & segmentation models by @roboflow 💜
> play with our real-time demo > fine-tune the models on your use case with our tutorials (takes a toaster's VRAM) > or just hand them to your agents 😄
RF-DETR just landed to @huggingface transformers 🥵🔥
sota real-time detection & segmentation models by @roboflow 💜
> play with our real-time demo > fine-tune the models on your use case with our tutorials (takes a toaster's VRAM) > or just hand them to your agents 😄
@skalskip92 @huggingface It's running detection independently on every frame with no tracking. Adding a tracker (ByteTrack) on top smooths it out.
@skalskip92 @huggingface Looking good!
@skalskip92 @huggingface this is my superbowl thank you x100
@skalskip92 @huggingface Man that looks clean, love it.
@mervenoyann @roboflow 💟 @roboflow
@ElBianco3 @huggingface soon
@skalskip92 @huggingface what about pose detection?
@DataDeLaurier @mervenoyann @Prince_Canuma @huggingface @roboflow This one lil segmentation model is gonna make me so much money hehehehe
@mervenoyann @Prince_Canuma @huggingface @roboflow "just landed" ???
wut...
@skalskip92 @huggingface Thats great! Can it be converted to NCNN? The nano version would be bonkers for single-board computers.
@measure_plan @mervenoyann @skalskip92 @huggingface We actually already support it! 🤗
You can find sample code in the model card of either the nano or medium variant: - https://huggingface.co/onnx-community/rfdetr_nano-ONNX - https://huggingface.co/onnx-community/rfdetr_medium-ONNX
But now with the recent transformers update, there are many more optimizations we can now make! Stay tuned.
@dreamworks2050 @mervenoyann @Prince_Canuma @huggingface @roboflow i use it everyday i was questioning the timing because it's been out for a year.
@xenovacom @mervenoyann @skalskip92 @huggingface oo very cool, i thought only object detection (not segmentation) was supported in js
will try it! thank you as always, you're amazing
@mervenoyann @huggingface @roboflow And it's already available on http://coreviz.io :)
@skalskip92 @diegopullano @huggingface What about real-time edge device applicability. Size of the http://best.pt as compared to YOLO variants
@skalskip92 @huggingface I tested RF-DETR a month ago on OpenVINO and TensorRT, and it was not this stable with the bounding boxes. Did something improve?
@skalskip92 @huggingface how does it compare to yolo26?
