Official Models

Domain Suite and Extension Packages

Computer Vision

Image Classification（backbone）

The following results are tested on Ascend 910 with ImageNet-1K.

model	acc@1	mindcv recipe	vanilla mindspore
vgg11	71.86	config
vgg13	72.87	config
vgg16	74.61	config	link
vgg19	75.21	config	link
resnet18	70.21	config	link
resnet34	74.15	config	link
resnet50	76.69	config	link
resnet101	78.24	config	link
resnet152	78.72	config	link
resnetv2_50	76.90	config
resnetv2_101	78.48	config
dpn92	79.46	config
dpn98	79.94	config
dpn107	80.05	config
dpn131	80.07	config
densenet121	75.64	config
densenet161	79.09	config
densenet169	77.26	config
densenet201	78.14	config
seresnet18	71.81	config
seresnet34	75.36	config
seresnet50	78.31	config
seresnext26	77.18	config
seresnext50	78.71	config
skresnet18	73.09	config
skresnet34	76.71	config
skresnet50_32x4d	79.08	config
resnext50_32x4d	78.53	config
resnext101_32x4d	79.83	config
resnext101_64x4d	80.30	config
resnext152_64x4d	80.52	config
rexnet_x09	77.07	config
rexnet_x10	77.38	config
rexnet_x13	79.06	config
rexnet_x15	79.94	config
rexnet_x20	80.64	config
resnest50	80.81	config
resnest101	82.50	config
res2net50	79.35	config
res2net101	79.56	config
res2net50_v1b	80.32	config
res2net101_v1b	95.41	config
googlenet	72.68	config
inceptionv3	79.11	config	link
inceptionv4	80.88	config	link
mobilenet_v1_025	53.87	config
mobilenet_v1_050	65.94	config
mobilenet_v1_075	70.44	config
mobilenet_v1_100	72.95	config
mobilenet_v2_075	69.98	config
mobilenet_v2_100	72.27	config
mobilenet_v2_140	75.56	config
mobilenet_v3_small	68.10	config
mobilenet_v3_large	75.23	config	link
shufflenet_v1_g3_x0_5	57.05	config
shufflenet_v1_g3_x1_5	67.77	config	link
shufflenet_v2_x0_5	57.05	config
shufflenet_v2_x1_0	67.77	config	link
shufflenet_v2_x1_5	57.05	config
shufflenet_v2_x2_0	67.77	config
xception	79.01	config	link
ghostnet_50	66.03	config
ghostnet_100	73.78	config
ghostnet_130	75.50	config
nasnet_a_4x1056	73.65	config
mnasnet_0.5	68.07	config
mnasnet_0.75	71.81	config
mnasnet_1.0	74.28	config
mnasnet_1.4	76.01	config
efficientnet_b0	76.89	config	link
efficientnet_b1	78.95	config	link
efficientnet_b2	79.80		link
efficientnet_b3	80.50		link
efficientnet_v2	83.77		link
regnet_x_200mf	68.74	config
regnet_x_400mf	73.16	config
regnet_x_600mf	73.34	config
regnet_x_800mf	76.04	config
regnet_y_200mf	70.30	config
regnet_y_400mf	73.91	config
regnet_y_600mf	75.69	config
regnet_y_800mf	76.52	config
mixnet_s	75.52	config
mixnet_m	76.64	config
mixnet_l	78.73	config
hrnet_w32	80.64	config
hrnet_w48	81.19	config
bit_resnet50	76.81	config
bit_resnet50x3	80.63	config
bit_resnet101	77.93	config
repvgg_a0	72.19	config
repvgg_a1	74.19	config
repvgg_a2	76.63	config
repvgg_b0	74.99	config
repvgg_b1	78.81	config
repvgg_b2	79.29	config
repvgg_b3	80.46	config
repvgg_b1g2	78.03	config
repvgg_b1g4	77.64	config
repvgg_b2g4	78.80	config
repmlp_t224	76.71	config
convnext_tiny	81.91	config
convnext_small	83.40	config
convnext_base	83.32	config
vit_b_32_224	75.86	config	link
vit_l_16_224	76.34	config
vit_l_32_224	73.71	config
swintransformer_tiny	80.82	config	link
pvt_tiny	74.81	config
pvt_small	79.66	config
pvt_medium	81.82	config
pvt_large	81.75	config
pvt_v2_b0	71.50	config
pvt_v2_b1	78.91	config
pvt_v2_b2	81.99	config
pvt_v2_b3	82.84	config
pvt_v2_b4	83.14	config
pit_ti	72.96	config
pit_xs	78.41	config
pit_s	80.56	config
pit_b	81.87	config
coat_lite_tiny	77.35	config
coat_lite_mini	78.51	config
coat_tiny	79.67	config
convit_tiny	73.66	config
convit_tiny_plus	77.00	config
convit_small	81.63	config
convit_small_plus	81.80	config
convit_base	82.10	config
convit_base_plus	81.96	config
crossvit_9	73.56	config
crossvit_15	81.08	config
crossvit_18	81.93	config
mobilevit_xx_small	68.90	config
mobilevit_x_small	74.98	config
mobilevit_small	78.48	config
visformer_tiny	78.28	config
visformer_tiny_v2	78.82	config
visformer_small	81.76	config
visformer_small_v2	82.17	config
edgenext_xx_small	71.02	config
edgenext_x_small	75.14	config
edgenext_small	79.15	config
edgenext_base	82.24	config
poolformer_s12	77.33	config
xcit_tiny_12_p16	77.67	config

Object Detection

The following results are tested on Ascend 910 with COCO2017.

yolo

model	map	mindyolo recipe	vanilla mindspore
yolov8_n	37.2	config
yolov8_s	44.6	config
yolov8_m	50.5	config
yolov8_l	52.8	config
yolov8_x	53.7	config
yolov7_t	37.5	config
yolov7_l	50.8	config
yolov7_x	52.4	config
yolov5_n	27.3	config
yolov5_s	37.6	config	link
yolov5_m	44.9	config
yolov5_l	48.5	config
yolov5_x	50.5	config
yolov4_csp	45.4	config
yolov4_csp(silu)	45.8	config	link
yolov3_darknet53	45.5	config	link
yolox_n	24.1	config
yolox_t	33.3	config
yolox_s	40.7	config
yolox_m	46.7	config
yolox_l	49.2	config
yolox_x	51.6	config
yolox_darknet53	47.7	config

Classic

model	map	mind_series recipe	vanilla mindspore
ssd_vgg16	23.2		link
ssd_mobilenetv1	22.0		link
ssd_mobilenetv2	29.1		link
ssd_resnet50	34.3		link
fasterrcnn	58	link	link
maskrcnn_mobilenetv1	coming soon		link
maskrcnn_resnet50	coming soon		link

Semantic Segmentation

model	mind_series recipe	vanilla mindspore
ocrnet	link	link
deeplab v3		link
deeplab v3 plus		link
unet		link

OCR

Text Detection

model	dataset	fscore	mindocr recipe	vanilla mindspore
dbnet_mobilenetv3	icdar2015	77.23	config	link
dbnet_resnet18	icdar2015	81.73	config	link
dbnet_resnet50	icdar2015	85.05	config	link
dbnet++_resnet50	icdar2015	86.74	config
psenet_resnet152	icdar2015	82.06	config	link
east_resnet50	icdar2015	84.87	config	link
fcenet_resnet50	icdar2015	84.12	config

Text Recognition

model	dataset	acc	mindocr recipe	vanilla mindspore
svtr_tiny	IC03,13,15,IIIT,etc	89.02	config
crnn_vgg7	IC03,13,15,IIIT,etc	82.03	config	link
crnn_resnet34_vd	IC03,13,15,IIIT,etc	84.45	config
rare_resnet34_vd	IC03,13,15,IIIT,etc	85.19	config	link

Text Orientation Classification

model	dataset	acc	mindocr recipe
mobilenetv3	RCTW17,MTWI,LSVT	94.59	config

Face

model	dataset	acc	mindface recipe	vanilla mindspore
arcface_mobilefacenet-0.45g	MS1MV2	98.70	config
arcface_r50	MS1MV2	99.76	config
arcface_r100	MS1MV2	99.38	config	link
arcface_vit_t	MS1MV2	99.71	config
arcface_vit_s	MS1MV2	99.76	config
arcface_vit_b	MS1MV2	99.81	config
arcface_vit_l	MS1MV2	99.75	config
retinaface_mobilenet_0.25	WiderFace	90.77/88.2/74.76	config	link
retinaface_r50	WiderFace	95.07/93.61/84.84	config	link

Language Models

model	mindformer recipe	vanilla mindspore
bert_base	config	link
t5_small	config
gpt2_small	config
gpt2_13b	config
gpt2_52b	config
pangu_alpha	config
glm_6b	config
glm_6b_lora	config
llama_7b	config
llama_13b	config
llama_65b	config
llama_7b_lora	config
bloom_560m	config
bloom_7.1b	config
bloom_65b	config
bloom_176b	config

Reinforcement Learning

Algorithm	Discrete Action Space	Continuous Action Space	CPU	GPU	Ascend	Environment	Reward
DQN	✅	/	✅	✅	✅	CartPole-v0	195
PPO	/	✅	✅	✅	✅	HalfCheetah-v2	4800
AC	✅	/	✅	✅	✅	CartPole-v0	195
A2C	✅	/	✅	✅	✅	CartPole-v0	195
DDPG	/	✅	✅	✅	✅	HalfCheetah-v2	4800
QMIX	✅	/	✅	✅	✅	SMAC/Simple Spread	90%/-145
SAC	/	✅	✅	✅	✅	HalfCheetah-v2	4800
TD3	/	✅	✅	✅	✅	HalfCheetah-v2	4800
C51	✅	/	✅	✅	✅	CartPole-v0	195
A3C	✅	/	/	✅	✅	CartPole-v0	195
CQL	/	✅	✅	✅	✅	Hopper-v0	3500
MAPPO	✅	/	✅	✅	✅	Simple Spread	-145
GAIL	/	✅	✅	✅	✅	HalfCheetah-v2	4800
AWAC	/	✅	✅	✅	✅	Ant-v2	5000
Dreamer	/	✅	/	✅	✅	Walker-walk	900
IQL	/	✅	✅	✅	✅	Walker2d-v2	3000
MADDPG	✅	/	✅	✅	✅	simple_spread	-140
Double DQN	✅	/	✅	✅	✅	CartPole-v0	195
Policy Gradient	✅	/	✅	✅	✅	CartPole-v0	195
Dueling DQN	✅	/	✅	✅	✅	CartPole-v0	195

Scientific Computing

Fields	Network	MindSpore Implementation	Ascend	GPU
General Physics	deepbsde	Link		✅
General Physics	pfnn	Link		✅
General Physics	pinns	Link		✅
Marine Physics	ocean_model	Link		✅
Electromagnetic	PINNs with Incremental Learning (incremental_learning)	Link	✅	✅
Electromagnetic	pinn_fwi		✅	✅
Electromagnetic	time_domain_maxwell	Link	✅	✅
Electromagnetic	frequency_domain_maxwell	Link	✅	✅
Electromagnetic	AD_FDTD	Link	✅	✅
Electromagnetic	SED_ANN	Link	✅	✅
Electromagnetic	Metasurface_holograms	Link	✅	✅
Computational Biology	MEGA-Fold	Link	✅	✅
Computational Biology	MEGA-EvoGen	Link	✅	✅
Computational Biology	MEGA-Assessment	Link	✅	✅
Computational Biology	ColabDesign	Link	✅	✅
Computational Biology	DeepDR	Link	✅	✅
Computational Biology	DeepFRI	Link	✅	✅
Computational Biology	FAAST	Link	✅	✅
Computational Biology	JT-VAE	Link	✅	✅
Computational Biology	KGNN	Link	✅	✅
Computational Biology	MG-BERT	Link	✅	✅
Computational Biology	Multimer	Link	✅	✅
Computational Biology	ProteinMPNN	Link	✅	✅
Computational Biology	UFold	Link	✅	✅
Computational Biology	ESM-IF1	Link	✅	✅
Computational Biology	ESM2	Link	✅	✅
Computational Biology	Grover	Link	✅	✅
Computational Biology	Pafnucy	Link	✅	✅
Computational Biology	SchNet	Link	✅	✅
Computational Biology	MolCT	Link	✅	✅
Computational Biology	PhysNet	Link	✅	✅
Computational Fluid Dynamics	FNO1D	Link	✅	✅
Computational Fluid Dynamics	KNO1D	Link	✅	✅
Computational Fluid Dynamics	FNO2D	Link	✅	✅
Computational Fluid Dynamics	KNO2D	Link	✅	✅
Computational Fluid Dynamics	FNO3D	Link	✅	✅
Computational Fluid Dynamics	CAE-LSTM		✅	✅
Computational Fluid Dynamics	eHDNN	Link	✅	✅
Computational Fluid Dynamics	HDNN	Link	✅	✅
Computational Fluid Dynamics	ViT	Link	✅	✅
Computational Fluid Dynamics	PeRCNN		✅	✅
Computational Fluid Dynamics	Burgers1D	Link	✅	✅
Computational Fluid Dynamics	Cylinder Flow	Link	✅	✅
Computational Fluid Dynamics	PDE-Net	Link	✅	✅

Foundation Model

Transformers

model	dataset	metric	score	mindformers config
bert_base_uncased	wiki	-	-	config
bert_tiny_uncased	wiki	-	-	config
qa_bert_base_uncased	SQuAD v1.1	EM / F1	80.74 / 88.33	config
tokcls_bert_base_chinese_cluener	CLUENER	Entity F1	0.7905	config
txtcls_bert_base_uncased_mnli	Mnli	Entity F1	84.80%	config
clip_vit_b_32	Cifar100	Accuracy	57.24%	config
clip_vit_b_16	Cifar100	Accuracy	61.41%	config
clip_vit_l_14	Cifar100	Accuracy	69.67%	config
clip_vit_l_14@336	Cifar100	Accuracy	68.19%	config
filip_vit_l_14	-	-	-	config
glm_6b	ADGEN	BLEU-4 / Rouge-1 / Rouge-2 / Rouge-l	8.42 / 31.75 / 7.98 / 25.28	config
gpt2	wikitext-2	-	-	config
gpt2_13b	wikitext-2	-	-	config
gpt2_52b	wikitext-2	-	-	config
llama_7b	alpac	-	-	config
llama_13b	alpac	-	-	config
llama_65b	alpac	-	-	config
mae_vit_base_p16	ImageNet-1K	-	-	config
vit_base_p16	ImageNet-1K	Accuracy	83.71%	config
pangualpha_2_6b	WuDaoCorpora	-	-	config
pangualpha_13b	WuDaoCorpora	-	-	config
swin_base_p4w7	ImageNet-1K	Accuracy	83.44%	config
t5_small	WMT16	-	-	config
t5_tiny	WMT16	-	-	config

Recommender

model	dataset	auc	mindrec recipe	vanilla mindspore
Wide&Deep	Criteo	0.8	link	link
Deep&Cross Network (DCN)	Criteo	0.8	link	link