Detection Model Zoo¶

Inference default use V100 16G.

YOLOX-PAI¶

Pretrained on COCO2017 dataset. (The result has been optimized with PAI-Blade, and only computes the model inference time. To learn about end2end inference time, you can refer to export.md.)

Algorithm	Config	Params	Speed^{V100 _{fp16 b32}}	mAP^{val _0.5:0.95}	AP^val ₅₀	Download
YOLOX-s	yolox_s_8xb16_300e_coco	9M	0.68ms	40.0	58.9	model - log
PAI-YOLOXs	yoloxs_pai_8xb16_300e_coco	16M	0.71ms	41.4	60.0	model - log
PAI-YOLOXs-ASFF	yoloxs_pai_asff_8xb16_300e_coco	21M	0.87ms	42.8	61.8	model - log
PAI-YOLOXs-ASFF-TOOD3	yoloxs_pai_asff_tood3_8xb16_300e_coco	24M	1.15ms	43.9	62.1	model - log
YOLOX-m	yolox_m_8xb16_300e_coco	25M	1.52ms	46.3	64.9	model - log
YOLOX-l	yolox_l_8xb8_300e_coco	54M	2.47ms	48.9	67.5	model - log
YOLOX-x	yolox_x_8xb8_300e_coco	99M	4.74ms	50.9	69.2	model - log
YOLOX-tiny	yolox_tiny_8xb16_300e_coco	5M	0.28ms	31.5	49.2	model - log
YOLOX-nano	yolox_nano_8xb16_300e_coco	2.2M	0.19ms	26.5	42.6	model - log

ViTDet¶

| Algorithm | Config | Params
(backbone/total) | Train memory
(GB) | inference time(V100)
(ms/img) | bbox_mAP^{val
_0.5:0.95 | mask_mAP^{val
_0.5:0.95 | Download |
| ———- | ———————————————————— | ———————— | ———————————————————— | ———————————————————— | ———————————————————— | ———————————————————— | ———————————————————— |
| ViTDet_MaskRCNN | vitdet_maskrcnn | 86M/111M | 13.3 (fp16) | 138ms | 50.65 | 45.41 | model - log |}}

FCOS¶

Algorithm	Config	Params (backbone/total)	Train memory (GB)	inference time(V100) (ms/img)	mAP^{val _0.5:0.95}	AP^val ₅₀	Download
FCOS-r50(caffe)	fcos-r50	23M/32M	5.0	85.8ms	38.58	57.18	model - log
FCOS-r50(torch)	fcos-r50	23M/32M	4.0 (fp16)	105.3ms	38.88	58.01	model - log

DETR¶

Algorithm	Config	Params (backbone/total)	Train memory (GB)	inference time(V100) (ms/img)	bbox_mAP^{val _0.5:0.95}	AP^val ₅₀	Download
DETR-r50	detr-r50	23M/41M	8.5	48.5ms	39.92	60.52	model - log
DAB-DETR-r50	dab-detr-r50	23M/43M	2.6	58.5ms	42.52	63.03	model - log
DN-DETR-r50	dab-detr-r50	23M/43M	7.8	58.5ms	44.39	64.66	model - log

DINO¶

Algorithm	Config	Params (backbone/total)	inference time(V100) (ms/img)	bbox_mAP^{val _0.5:0.95}	AP^val ₅₀	Download	Comment
DINO_4sc_r50_12e	DINO_4sc_r50_12e	23M/47M	184ms	48.71	66.27	model - log	Inference use V100 32G
DINO_4sc_r50_36e	DINO_4sc_r50_36e	23M/47M	184ms	50.69	68.60	model - log	Inference use V100 32G
DINO_4sc_swinl_12e	DINO_4sc_swinl_12e	195M/217M	155ms	56.86	75.61	model - log	Inference use V100 32G
DINO_4sc_swinl_36e	DINO_4sc_swinl_36e	195M/217M	155ms	58.04	76.76	model - log	Inference use V100 32G
DINO_5sc_swinl_36e	DINO_5sc_swinl_36e	195M/217M	235ms	58.47	77.10	model - log	Inference use V100 32G
DINO++_5sc_swinl_18e	DINO++_5sc_swinl_18e	195M/218M	325ms	63.39	80.25	model - log	Inference use A100 80G