Available architectures

Here is the complete list of all the neural network architectures available in Studio. When available, links to the research papers are provided.

Pre-training

All our architectures are pre-trained on real-world, open datasets. Training them from scratch would take days or weeks to have correct predictions. Thanks to the pre-training, all you are doing is fine-tuning the model to your specific use-case. You can see the pre-training dataset in the table below.

Classification and tagging backbone architectures

Backbone	Inference time on Nvidia T4 (ms) (Sync)	Inference time on Nvidia T4 (ms) (ASync)	Input size	Training batch size	Trained on	Research paper
EfficientNet B0	25.0	15.9	224x224	32	ImageNet	arXiv
EfficientNet B1	27.6	17.5	240x240	32	ImageNet	arXiv
EfficientNet B2	28.5	18.2	260x260	32	ImageNet	arXiv
EfficientNet B3	30.6	19.6	300x300	32	ImageNet	arXiv
EfficientNet B4	35.2	23.2	380x380	16	ImageNet	arXiv
EfficientNet B5	47.8	35.6	456x456	8	ImageNet	arXiv
EfficientNet B6	66.8	55.6	528x528	4	ImageNet	arXiv
Inception-Resnet v2	38.8	28.2	299x299	32	ImageNet	arXiv
ResNet-50	28.6	17.8	224x224	32	ImageNet	arXiv
ResNet-101	33.8	21.2	224x224	32	ImageNet	arXiv
ResNet-152	37.0	24.6	224x224	32	ImageNet	arXiv
Inception v1	22.8	15.0	224x224	32	ImageNet	arXiv
Inception v2	25.4	15.9	224x224	32	ImageNet	arXiv
Inception v3	28.4	20.2	299x299	32	ImageNet	arXiv
Inception v4	34.8	25.5	299x299	32	ImageNet	arXiv
VGG 16	78.3	67.0	224x224	32	ImageNet	arXiv

Detection meta-architectures

Architecture	Backbone	Inference time on Nvidia T4 (ms) (Sync)	Inference time on Nvidia T4 (ms) (ASync)	Input size	Training batch size	Trained on	Research paper
EfficientDet	Eff. Net B0	82.6	49.2	512x512	16	COCO 2018	arXiv
	Eff. Net B1	117	69.3	640x640	8	COCO 2018	arXiv
	Eff. Net B2	158	96.3	768x768	4	COCO 2018	arXiv
	Eff. Net B3	221	132	896x896	2	COCO 2018	arXiv
	Eff. Net B4	316	173	1024x1024	1	COCO 2018	arXiv
Yolo v3	Darknet 53	56.7	36.0	416x416	64	ImageNet 2012	arXiv
Yolo v8	Nano	44.3	28.6	640x640	16	COCO 2018	arXiv
	Small	4.74	31.6	640x640	16	COCO 2018	arXiv
	Medium	57.4	45.0	640x640	16	COCO 2018	arXiv
	Large	75.5	64.3	640x640	8	COCO 2018	arXiv
	Extra	99.3	87.7	640x640	8	COCO 2018	arXiv
Faster-RCNN	Resnet-50	N/A	N/A	1024x1024*	1	COCO 2018	arXiv
	Resnet-101	174	169	1024x1024*	1	COCO 2018	arXiv
SSD	Inception v2	34.8	17.0	300x300	24	COCO 2018	arXiv
	MobileNet v1	32.5	16.4	300x300	24	COCO 2018	arXiv
	MobileNet v2	32.4	18.1	300x300	24	COCO 2018	arXiv
SSDLite	MobileNet v2	28.7	16.2	300x300	24	COCO 2018	arXiv

(*) Faster-RCNN does not require a fixed image input size. As such they can accept images from 600 to 1024 pixel.