Training Resolutions by Model Type

Training resolution affects model accuracy, inference speed, and training time. Each model architecture has a default resolution that balances these factors. By default, Roboflow suggests the default training resolution for the selected model architecture.

The table below shows the default training resolution for each model architecture and size. You can override these defaults by configuring the resize preprocessing step when creating a new Dataset Version.

Object Detection

Model TypeFamily & SizeDefault Training Resolution
Object DetectionRF-DETR Nano384x384
Object DetectionRF-DETR Small512x512
Object DetectionRF-DETR Medium576x576
Object DetectionRF-DETR Large704x704
Object DetectionRF-DETR X Large700x700
Object DetectionRF-DETR 2X Large880x880
Object DetectionRoboflow 3.0 - Fast640x640
Object DetectionRoboflow 3.0 - Accurate640x640
Object DetectionRoboflow 3.0 - Medium640x640
Object DetectionRoboflow 3.0 - Large640x640
Object DetectionRoboflow 3.0 - Extra Large640x640
Object DetectionYOLOv26 (n/s/m/l/x)640x640
Object DetectionYOLOv12 (n/s/m/l/x)640x640
Object DetectionYOLOv11 (n/s/m/l/x)640x640
Object DetectionYOLOv10 (n/s/m/b/l/x)640x640
Object DetectionYOLOv9 (s/m/c/e)640x640
Object DetectionYOLOv8 (n/s/m/l/x)640x640
Object DetectionYOLOv5 (n/s/m/l/x)640x640
Object DetectionYOLOv7 (legacy)640x640
Object DetectionYOLO-NAS Small640x640
Object DetectionYOLO-NAS Medium640x640
Object DetectionRoboflow Instant1008x1008

Instance Segmentation

Model TypeFamily & SizeDefault Training Resolution
Instance SegmentationRF-DETR Seg Nano312x312
Instance SegmentationRF-DETR Seg Small384x384
Instance SegmentationRF-DETR Seg Medium432x432
Instance SegmentationRF-DETR Seg Large504x504
Instance SegmentationRF-DETR Seg X Large624x624
Instance SegmentationRF-DETR Seg 2X Large768x768
Instance SegmentationRoboflow 3.0 - Fast (Seg)640x640
Instance SegmentationRoboflow 3.0 - Accurate (Seg)640x640
Instance SegmentationRoboflow 3.0 - Medium (Seg)640x640
Instance SegmentationRoboflow 3.0 - Large (Seg)640x640
Instance SegmentationRoboflow 3.0 - Extra Large (Seg)640x640
Instance SegmentationYOLO-seg (v8/10/11/12)640x640
Instance SegmentationSAM3 (Segment Anything 3)1008x1008
Instance SegmentationSemantic segmentation (DeepLabV3+)>= 512x512

Classification & Pose

Model TypeFamily & SizeDefault Training Resolution
Classification & PoseResnet-18/34/50224x224
Classification & PoseYOLO-cls (v8/11)224x224
Classification & PoseVision Transformer (ViT)224x224
Classification & PoseYOLO-pose (keypoints)640x640

Multimodal/VLM

Model TypeFamily & SizeDefault Training Resolution
Multimodal/VLMPaliGemma 2 - 3 B448x448
Multimodal/VLMPaliGemma 2 - 10 B/28 B448x448
Multimodal/VLMFlorence-2448x448
Multimodal/VLMQWEN 2.5 VL448x448
Multimodal/VLMSmolVLM2384x384