Test photo drone-city-block

drone-city-block

License: CC0-1.0
Author: Kleidieski
Source: Wikimedia Commons, https://commons.wikimedia.org/wiki/File:Former_Enver_Hoxha_House_territory_(drone_view).jpg (CC0 1.0 by the author, checked 2026-10-27); downscaled to 1600x900, metadata removed
Hard cases: top-down, small-objects

Hover any output to magnify the same spot in every output of that task.

Object detection

D-FINE l — D-FINE lFails: boxes a poolside feature as a clock

D-FINE m

D-FINE n — D-FINE nFails: boxes a rooftop as a traffic light

D-FINE s — D-FINE sFails: boxes a garden fountain as a clock

D-FINE x — D-FINE xFails: boxes a rooftop as a kite

DEIMv2 l

DEIMv2 m — DEIMv2 mFails: boxes a roof terrace as a clock

DEIMv2 s

DEIMv2 x

Grounding DINO base

Grounding DINO tiny

GTR l

GTR m

GTR s

GTR x

LLMDet base

LLMDet large

LLMDet tiny

MM-Grounding-DINO base

MM-Grounding-DINO large

MM-Grounding-DINO tiny

RF-DETR large

RF-DETR medium

RF-DETR nano

RF-DETR small

RT-DETRv4 l

RT-DETRv4 m

RT-DETRv4 s — RT-DETRv4 sFails: boxes the oval garden bed as a clock

RT-DETRv4 x

YOLO11 l — YOLO11 lFails: boxes a whole building as a truck

YOLO11 m

YOLO11 n

YOLO11 s

YOLO11 x — YOLO11 xFails: boxes a paved courtyard as a clock

YOLO26 l

YOLO26 m

YOLO26 n

YOLO26 s

YOLO26 x

YOLOv8 l

YOLOv8 m

YOLOv8 n

YOLOv8 s

YOLOv8 x

Instance segmentation

EoMT large-1280 — EoMT large-1280Fails: masks merge whole rows of parked cars

EoMT large-640

GTR-Seg l

GTR-Seg m

GTR-Seg s — GTR-Seg sFails: carrot mask on a patch of trees

GTR-Seg x

RF-DETR-Seg 2xlarge

RF-DETR-Seg large

RF-DETR-Seg medium

RF-DETR-Seg nano

RF-DETR-Seg small

RF-DETR-Seg xlarge

YOLO11-seg l — YOLO11-seg lFails: truck mask on a whole building

YOLO11-seg m

YOLO11-seg n

YOLO11-seg s

YOLO11-seg x — YOLO11-seg xFails: clock mask on a paved courtyard

YOLO26-seg l

YOLO26-seg m — YOLO26-seg mFails: car mask on a rooftop

YOLO26-seg n

YOLO26-seg s

YOLO26-seg x — YOLO26-seg xFails: clock mask on a rooftop

YOLOv8-seg l — YOLOv8-seg lFails: clock mask on a paved courtyard

YOLOv8-seg m

YOLOv8-seg n

YOLOv8-seg s

YOLOv8-seg x

Depth estimation

Depth Anything 3 Metric large — Depth Anything 3 Metric largeFails: top-down view melts into a blurred bowl; buildings barely separate

Depth Anything 3 Mono large

Depth Anything V2 base

Depth Anything V2 large

Depth Anything V2 small

MoGe-2 vitb

MoGe-2 vitl

MoGe-2 vits