Runs as a custom pluginDetectionRecommendedVerified 2026-11-05
D-FINE L (COCO)
ustc-community/dfine-large-coco
D-FINE L (COCO) is a image object detection model (Apache-2.0). Can run on Mixpeek as a custom plugin.
Mixpeek does not run it today. It can run as a custom plugin: the weights load without remote code.
No demo for D-FINE L (COCO) yet
A video, image, slideshow or playground goes here once we record one on our own data.
What it is good and bad at
- Best at
- Real-time DETR with fine-grained box refinement; among the highest COCO AP per latency of the permissive detectors.
- Worst at
- Closed COCO label set.
- Task
- Detection
- Input
- Image
- Output
- Boxes, Labels, Scores
- Use cases
- Object detection
- Why we recommend it
- Apache-2.0 detector with safetensors and transformers support: the permissive choice over AGPL YOLO. Not yet evaluated by Mixpeek.
Templates that use it
No template uses this model yet. Browse templates.
Input and output
Input
- image
- resolution
- 640x640
- color
- RGB
- prompts
- none (fixed classes)
Output
- boxes
- xyxy pixels (post_process_object_detection)
- labels
- COCO 80 classes
- scores
- 0..1
How Mixpeek wires it
- Extractors
- None today
- Feature URIs
- None today
- Vector indexes
- None
- Retriever stages
- None
- Combines with
- detector boxes -> crops -> SigLIP vectors per object (one document per object, or a multi-vector per frame)
- labels -> payload field -> attribute filter in a retriever
- scene splits -> keyframes -> detector -> object documents with frame timestamps
How it runs
- Status
- Runs as a custom plugin Not served. Permissive alternative to Ultralytics YOLO. Would run as a custom extractor plugin loading safetensors with transformers (trust_remote_code false), emitting boxes, labels and crops.
- Serve apps
- None
- Runtime
- transformers DFineForObjectDetection + RTDetrImageProcessor
- Weights
- safetensors, no remote code
- Parameters
- 31 M
- Size on disk
- 126 MB
- Preprocessing
- Resize to 640x640, rescale to 0..1 (image processor).
- Postprocessing
- Sigmoid class scores, top-k queries, threshold; no NMS (one-to-one matching).
- Serving
- both
- Hardware
- Runs on CPU (slower without a GPU)
Scale to zero
- Minimum replicas
- 0
- Maximum replicas
- 1
- Idle timeout
- 600 s
- Memory
- 2 GB
- CPUs
- 2
- GPUs
- 0
- Cold start, node warm
- Not recorded
- Cold start, new node
- Not recorded
Measured cost
Unmeasured
No committed measurement; plan row 3 measures it in the standalone image.
Our evaluations
Unmeasured
We have not evaluated this model on our own data yet. Publisher results are in the sources below.
Questions
- What is D-FINE L (COCO) best at?
- Real-time DETR with fine-grained box refinement; among the highest COCO AP per latency of the permissive detectors.
- What are the limits of D-FINE L (COCO)?
- Closed COCO label set.
- Can D-FINE L (COCO) be used commercially?
- Yes. Apache-2.0 weights and code; no use restrictions.
- How does D-FINE L (COCO) run on Mixpeek?
- Not served. Permissive alternative to Ultralytics YOLO. Would run as a custom extractor plugin loading safetensors with transformers (trust_remote_code false), emitting boxes, labels and crops.