欢迎使用Ascend ACL_PyTorch

为方便更多开发者体验和使用昇腾芯片澎湃推理算力，该目录下提供了经典和主流算法模型实现昇腾服务器推理的端到端流程，更多模型持续更新中。如果您有任何需求，请在modelzoo/issues提交issue，我们会及时处理。如果您希望适配您的自定义模型，我们同时提供在昇腾设备上自行适配模型的ONNX推理流程。

如何贡献

在开始贡献之前，请先阅读CONTRIBUTING。谢谢！

目前ACL_PyTorch仓库已有模型398个

注意：
在提交新模型时，请加上模型ID用于区分，为防止重复提交模型，请执行脚本get_modelID.py，该脚本会自动检索ACL_PyTorch仓库中所有与您提交模型相关的已有模型，请自行查看脚本给出的链接，如果均不同，则可以输入1或true用于获取模型ID。由于该脚本使用正则匹配，后续新模型刷新到主页需要添加README内容时，格式请参考其余模型，并且同步刷新上文模型数量。脚本执行方式如下：
python3 get_modelID.py --model your_model_name

参数说明：
--model：请输入您所需提交新模型的简称，比如Conformer-base模型，您可以输入conformer用于检索所有相关模型（大小写不敏感）

支持模型列表（按字母顺序排序）

说明：
以下无精度指标的模型均需人工与在线推理结果比较
因使用版本差异，模型性能可能存在波动，性能仅供参考

规范模型

CV-classfication

ID	Name	Dataset	精度			300I Pro最优性能(对应bs)	输入shape
ID	Name	Dataset	Top1Acc	Top5Acc	mAP	300I Pro最优性能(对应bs)	输入shape
100007	Big-Transfer	CIFAR-10	97.62%			1758.00(bs16)	bs x 3 x 128 x 128
100008	C3D	UCF101	81.87%			54.92(bs4)	bs x 10 x 3 x16 x 112 x 112
100019	DenseNet121	ImageNet	74.43%	91.96%		2195(bs8)	bs x 3 x 224 x 224
100033	GaitSet	CASIA-B	95.512%			723(bs64)	bs x 100 x 64 x 44
100037	InceptionV3	ImageNet	77.31%	93.46%		2388(bs8)	bs x 3 x 299 x2 99
100038	MaskRcnn	coco			53.7149%	22.06(bs1)	1 x 3 x 1344 x 1344
100039	MobileNetV1	ImageNet	69.52%	89.05%		13990(bs32)	bs x 3 x 224 x 224
100041	MobileNetV2	ImageNet	71.87%	90.32%		7072(bs4)	bs x 3 x 224 x 224
100040	MobileNetV3	ImageNet	65.094%	85.432%		15442.12(bs32)	bs x 3 x 224 x 224
100075	RegNetX-1.6GF	ImageNet	76.93%	93.43%		5426.759(bs8)	bs x 3 x 224 x 224
100044	RepVGG	ImageNet	72.15%	90.4%		8929(bs32)	bs x 3 x 224 x 224
100076	ResNeSt50	ImageNet	80.98%			1704(bs4)	bs x 3 x 224 x 224
100077	ResNext101-32x8d	ImageNet	79.312%	94.526%		1251(bs8)	bs x 3 x 224 x 224
100078	SCNet	ImageNet	80.34%			2135(bs4)	bs x 3 x 224 x 224
100053	ShuffleNetV1	ImageNet	67.71%			7847(bs16)	bs x 3 x 224 x 224
100052	ShuffleNetV2	ImageNet	69.33%	88.34%		7736(bs32)	bs x 3 x 224 x 224
100079	SqueezeNet1	ImageNet	57.32%	80.06%		21301(bs8)	bs x 3 x 224 x 224
100068	VGG16	ImageNet	71.28%	90.38%		1424(bs16)	bs x 3 x 224 x 224
100070	ViT	ImageNet	80.63%(patch32_224)			1679.63(patch32_224 bs64)	bs x 3 x 224 x 224
100401	Chinese-CLIP					33ms(img,bs20)	20 x 3 x 224 x 224

CV-detection

ID	Name	Dataset	精度			300I Pro最优性能(对应bs)	800I A2最优性能(对应bs)	输入shape
ID	Name	Dataset	AP	mAP	Acc	300I Pro最优性能(对应bs)	800I A2最优性能(对应bs)	输入shape
100080	CascadeRCNN-DCN	coco		43.8%		3.9(bs1)		1 x 3 x 1216 x 1216
100010	CenterFace	WIDER_FACE		hard:74.55% easy:92.24% Medium:91.02%		439.9(bs1)		bs x 3 x 800 x 800
100011	CenterNet	coco	36.4%			34.1(bs4)		bs x 3 x 512 x 512 bs x 3 x 800 x 800
100012	CRNN-BuildIn	IIIT5K_lmdb	74.87%			17815(bs64)		bs x 1 x 32 x 100
100083	DBNet-MobileNetV3	ICDAR2015			77.5%	196(bs1)		bs x 736 x 1280 x 3
100020	DETR	coco	41.6%			63.75(bs1)		多尺度
100023	EfficientDet-D0	coco	33.4%			260(bs4)		bs x 3 x 512 x 512
100024	EfficientNet-B0	ImageNet			75.088%(top1)	2489(bs8)		bs x 3 x 224 x 224
100042	OpenPose	coco	40.4%			887.45(bs4)		bs x 3 x 368 x 6406
100043	PSENet	ICDAR2015			acc:80.5% recall:63.9%	70(bs1)		bs x 3 x 704 x 1216
100084	RetinaNet-r50-fpn	coco		36.3%		15.48(bs1)		1 x 3 x 1216 x 1216
100048	RetinaNet-ResNet18	coco		31.6%		50(bs1)		1 x 3 x 1216 x 1216
100057	SSD	coco		25.4%		337.01(bs4)		bs x 3 x 300 x 300
100056	SSD-ResNet34	coco		23%		1324(bs8)		bs x 3 x 300 x 300
100085	YOLOV3	coco		63.30%		219(bs4)		bs x 3 x 640 x 640
100072	YOLOV4	coco		60.3%		171.15(bs8)		bs x 3 x 416 x 416
100073	YOLOV5s2.0	coco		55.3%		998.004(bs4)		bs x 3 x 640 x 640
100074	YOLOV5s6.0	coco		55.9%		737.04(bs4)		bs x 3 x 640 x 640
100086	YOLOX	coco	51.2%			77.4(bs64)		bs x 3 x 640 x 640
100399	GLIP	coco	46.3%			0.62(bs1)		bs x 3 x H x W
100400	GLIP_STATIC					227.43ms(bs1)		1 x 3 x 784 x 1344
100404	InternImage_Detection	coco		box: 55.6% seg: 48.6%		1614ms(bs1)		1 x 3 x 1216 x 1216
100406	GroundingDINO	coco		52.4%		1378ms(bs1)	877ms(bs1)	多尺度

CV-segmentation

ID	Name	Dataset	精度		300I Pro最优性能（对应bs）	输入shape
ID	Name	Dataset	Acc	mIoU	300I Pro最优性能（对应bs）	输入shape
100016	DeeplabV3	Cityscapes	79.12%		4.9(bs1)	1 x 3 x 1024 x 2048
100055	SOLOV2	coco	34%		22(bs1)	1 x 3 x 800 x 1216
100088	UNet	carvana	98.6%		75(bs1)	bs x 3 x 572 x 572
100405	InternImage_Segmentation	ADE20K	aAcc: 86.55% mAcc: 71.8%	59.54%	2067ms(bs1)	多尺度

CV-gan

ID	Name	Dataset	精度	最优性能(对应bs)	输入shape
ID	Name	Dataset	精度	300I Pro	输入shape
100059	StarGAN	celeba		1186(bs8)	bs x 3 x 128 x 128 bs x 5
100061	StyleGAN2-ADA	代码仓提供		39.81(bs1)	1 x 512

CV-pose_estimation

ID	Name	Dataset	精度		300I Pro最优性能(对应bs)	输入shape
ID	Name	Dataset	Top1Acc	Top5Acc	300I Pro最优性能(对应bs)	输入shape
100036	HRNet	ImageNet	76.46%	93.14%	2036(bs16)	bs x 3 x 224 x 224
100060	STGCN	Kinetics	31.59%	53.74%	381(bs1)	bs x 3 x 300 x 18 x 2
100407	MuseTalk	代码仓提供			229s(8bs)	多尺度

CV-super_resolution

ID	Name	Dataset	精度	300I Pro最优性能(对应bs)	输入shape
ID	Name	Dataset	PSNR	300I Pro最优性能(对应bs)	输入shape
100022	EDSR	DIV2K	34.6	7.9(bs1)	bs x 3 x 1020 x 1020
100089	EDSR-Dynamic	B100	32.35	6.9(H:240, W:320)	多尺度

CV-tracking

ID	Name	Dataset	精度	300I Pro最优性能(对应bs)	输入shape
ID	Name	Dataset	Acc	300I Pro最优性能(对应bs)	输入shape
100017	Deepsort	MOT16	30.0%	yolov3:464.29(bs1) deep:2950(bs1)	bs x 3 x 416 x 416
100090	T2Vec	Proto	(精度数据参考链接)	9.85ms	动态输入

CV-image_registration

ID	Name	Dataset	精度		300I Pro最优性能(对应bs)	输入shape
ID	Name	Dataset	Auc@20	Precision	300I Pro最优性能(对应bs)	输入shape
100091	SuperGlue-SuperPoint	YFCC100M	74.72%	97.80%	1.4(e2e)	1 x 1 x 1200 x 1600

CV-video_understanding

ID	Name	Dataset	精度		300I Pro最优性能(对应bs)	输入shape
ID	Name	Dataset	Top1Acc	Top5Acc	300I Pro最优性能(对应bs)	输入shape
100092	TSM-SthV2	sthv2	61.87%	87.21%	20.77(bs1)	bs x 48 x 3 x 256 x 256

Audio

ID	Name	Dataset	精度			300I Pro最优性能(对应bs)	800I A2最优性能(对应bs)	输入shape
ID	Name	Dataset	WER	CER	Acc	300I Pro最优性能(对应bs)	800I A2最优性能(对应bs)	输入shape
100018	DeepSpeech2	an4	9.573	5.515		7.74(bs32)		bs x 1 x 161 x 621 bs x 1
100027	EspNet	代码仓提供				430(分档)		多尺度
100032	FastSpeech2	LJSpeech				13.66(bs1)		多尺度
100035	HiFiGAN	LJSpeech				637(bs8 mel_len:250)		多尺度
100063	Tacotron2	LJSpeech				33508(bs16)		多尺度
100064	TDNN-buildin	Mini Librispeech			99.93%	1358(bs4)		bs x 1800 x 24
100093	Conformer	aishell			95.04%	60		多尺度
100402	SenseVoice	代码仓提供				80ms(bs1)	40ms(bs1)	多尺度
100403	CosyVoice	代码仓提供				2.0s(rtf)	0.3s(rtf)	多尺度

Nlp

ID	Name	Dataset	精度	300I Pro最优性能(对应bs)	输入shape
ID	Name	Dataset	Acc	300I Pro最优性能(对应bs)	输入shape
100001	ALBERT	SST-2	92.8%	1350(bs16)	bs x 128
100094	BertBase-Cased-SST2	SST-2	92.43%	2906(bs32)	bs x 128 bs x 64
100003	Bert-Base-CH	zhwiki	77.94%	254(bs8)	bs x 384
100340	TextCNN	THUCNews	top1acc:90.47% top5acc:99.35%	29237(bs64)	bs x 32
100095	Bert-Uncased-Huggingface	SQuAD 1.1	F1:88.2% EM:80.84%	328.64(bs4)	bs x 384
100026	Ernie3	Clue	49%	1313(bs8)	bs x 128
100096	M2M100	sacrebleu		encoder:224.376(bs1) der_first_step:82.827(bs1) decoder:141.6(bs1)	1 x 90
100050	RoBERTa	SST-2	94.8%	1473(bs32)	bs x 70
100051	SAST	ICDAR	91.3%	22(bs1)	bs x 3 x 896 x 1536
100006	BiLSTM-CRF	CLUE_NER	acc=73.5% recall=69.3% f1=71.4%	961(bs32)	ids:bs,50;mask:bs,50
100348	Uie_for_PyTorch	doccano(paddle)	f1=100%	1334.26(bs16)	input_ids:bsx512;token_type_ids:bsx512;position_ids:bsx512;attention_mask:bsx512
100349	Pet_for_PyTorch	eprstmt	f1=88.5%	342.79(bs8)	bs x 128;bs x 128;bs x 128;bs x 1 x 1 x 128

OCR

ID	Name	Dataset	精度			最优性能（对应bs）	输入shape
ID	Name	Dataset	Top1Acc	Top5Acc	mAP	300I Pro	输入shape
100082	CRNN-ocr	原仓自带的数据集	78.37%			7969(bs64)	bs x 1 x 32 x 160
100087	DBNET-ocr	icdar2015			88%	19(bs16)	bs x 3 x 736 x 1280

生态贡献模型

CV-classfication

ID	Name	Dataset	精度			最优性能（对应bs）		输入shape
ID	Name	Dataset	Top1Acc	Top5Acc	mAP	300I Pro		输入shape
100098	3D-AttentionNet	CIFAR-10	62.2%			7806.96(bs16)		bs x 3 x 32 x 32
100099	3D-ResNets	hmdb51	62.22%			830.7165(bs10)		10 x 3 x 16 x 112 x 112
100100	AlexNet	ImageNet	56.56%	79.1%		12672(bs64)		bs x 3 x 224 x 224
100101	Dino-ResNet50-baseline	ImageNet	75.28%	92.56%		87537(bs64)		bs x 3 x 224 x 224
100102	BMN	Activity1.3	67.69%			114.34(bs1)		bs x 400 x 100
100103	BEIT	ImageNet	84.68%			516.00(bs8)		bs x 3 x 224 x 224
100104	CH-PPOCR-MobileNetV2.0	PaddleOCR				31958.773(bs64)		bs x 3 x 48 x 192
100105	Conformer-base	ImageNet	83.85%			257.7437(bs8)		bs x 3 x 224 x 224
100106	Conformer-Ti	ImageNet	81.09%			907.58(bs8)		bs x 3 x 224 x 224
100107	ConvMixer	ImageNet	81.37%			102.9(bs1)		bs x 3 x 224 x 224
100108	ConvNext	ImageNet	82.094%			461.9(bs8)		bs x 3 x 224 x 224
100109	CSPResNeXt50	ImageNet	79.79%			3251(bs4)		bs x 3 x 224 x 224
100110	CSWin-Transformer	ImageNet	83.3%			223.5(bs16)		bs x 3 x 224 x 224
100111	Deit-Small	ImageNet	79.5%	94.83%			415(bs1)	bs x 3 x 224 x 224
100112	DPN131	ImageNet	79.47%	94.54%		550(bs4)		bs x 3 x 224 x 224
100013	CRNN-Meijieru	demo文件				19374(bs64)	bs x 1 x 32 x 100
100113	Efficient-3DCNNs	UCF-101	81.073%	96.325%		1245.1167(bs4)		bs x 3 x 16 x 112 x 112
100114	EfficientNet-B1	ImageNet	75.940%	92.774%		1409.692(bs8)		bs x 3 x 240 x 240
100115	EfficientNet-B3	ImageNet	76.25%	92.56%		739.02(bs16)		bs x 3 x 300 x 300
100116	EfficientNet-B5	ImageNet	77.2%	92.8%		166.408(bs64)		bs x 3 x 456 x 456
100117	EfficientNet-B7	ImageNet	84.4%			75.9(bs32)		bs x 3 x 600 x 600
100025	EfficientNet-V2	ImageNet	82.26%			1670(bs64)		bs x 3 x 288 x 288
100118	FixRes	ImageNet	79.0%			984(bs4)		bs x 3 x 224 x 224
100119	FocalTransformer	ImageNet	83.586%			7.96(bs1)		bs x 3 x 224 x 224
100120	GENet	CIFAR-10	94.23%			9981(bs16)		bs x 3 x 32 x 32
100121	GhostNet1.0x	ImageNet	73.98%			4974(bs8)		bs x 3 x 224 x 224
100122	GloRe	UCF101	92.12%	99.56%		85(bs4)		bs x 3 x 8 x 224 x 224
100123	GoogleNet	ImageNet	69.78%	89.53%		6308.38(bs8)		bs x 3 x 8 x 224 x 224
100124	HRNet-Image-Classification	ImageNet	76.51%	93.22%		1986(bs16)		bs x 3 x 224 x 224
100125	InceptionResNetV2	ImageNet	80.15%	95.24%		1310.5(bs8)		bs x 3 x 299 x 299
100126	InceptionV4	ImageNet	79.99%	94.86%		1498.5(bs4)		bs x 3 x 299 x 299
100127	LResNet100E-IR	LFW	99.7%			1432(bs16)	746(bs16)	bs x 3 x 112 x 112
100128	LV-Vit	ImageNet	83.3%			407.14(bs8)		bs x 3 x 224 x 224
100129	MAE	ImageNet	83.52%%			266.8(bs1)		bs x 3 x 224 x 224
100130	MnasNet	ImageNet	73.48%			10650.988(bs16)		bs x 3 x 224 x 224
100131	MobileNetV3-large	ImageNet	75.62%	92.47%		6998.89(bs16)		bs x 3 x 224 x 224
100132	MOCOV2	ImageNet	67.28%	87.82%		3288.38(bs4)		bs x 3 x 224 x 224
100133	OSNet	Market-1501			82.55%	4075(bs8)		bs x 3 x 256 x 128
100134	PAMTRI	veri			68.64%	1564.274(bs4)		bs x 3 x 256 x 256
100135	PnasNet5large	ImageNet	81.76%			203.256(bs4)		bs x 3 x 331 x 331
100136	PointNet	shapenetcore	97.35%			2374.11(bs1)		bs x 3 x 2500
100137	PointNetCNN	modelnet40	82.82%			273(bs1)		1 x 1024 x 3
100138	PointNet+	modelnet40	88.4%			partone7825(bs4) parttwo5127(bs1)		partone[bs, 512, 3] [bs, 3, 32, 512] parttwo[bs, 3, 128] [bs, 131, 64, 128]
100139	R(2+1)D	UCF-101	89.23%	97.45%		84.7777(bs32)		bs x 3 x 224 x 224
100140	ReID	Market1501			85.9%	4417.628(bs16)		bs x 3 x 256 x 128
100141	RegNetY-1.6GF	ImageNet	77.86%	93.72%		4417.628(bs16)		bs x 3 x 224 x 224
100142	Res2Net101-v1b	ImageNet	81.22%	95.36%			347(bs32)	bs x 3 x 224 x 224
100143	ResNet152	ImageNet	78.31%	94.05%		1844(bs8)		bs x 3 x 256 x 256
100144	ResNet34	ImageNet	73.31%	91.44%		5455(bs16)		bs x 3 x 224 x 224
100145	ResNet50-MMLab	cifar100	79.9%			9329(bs16)		bs x 3 x 32 x 32
100045	ResNet18	ImageNet	69.75%	89.10%		9828(bs128)		bs x 3 x 256 x 256
100046	ResNet50	ImageNet	76.14%	92.87%		4250(bs64)		bs x 3 x 256 x 256
100146	ResNet50-mlperf	ImageNet	76.44%		3940.45(bs64)		bs x 3 x 256 x 256
100147	ResNet101	ImageNet	77.38%	93.56%		2548(bs8)		bs x 3 x 256 x 256
100148	ResNetvd	ImageNet	77.37%	93.77%		2823(bs32)		bs x 3 x 224 x 224
100047	ResNeXt50	ImageNet	77.61%			3749(bs32)		bs x 3 x 224 x 224
100149	SENet	ImageNet	77.64%	93.74%		2479(bs32)		bs x 3 x 224 x 224
100150	Se-ResNext101	ImageNet	78.24%			927.09(bs4)		bs x 3 x 224 x 224
100151	SE-ResNet50	ImageNet	77.36%	93.76%		2690(bs32)		bs x 3 x 224 x 224
100152	SE-ResNeXt50-32x4d	ImageNet	79.06%	94.44%		1804.86(bs4)		bs x 3 x 224 x 224
100153	ShiftViT	ImageNet	79.3%			842.25(bs8)		bs x 3 x 224 x 224
100154	ShuffleNetv2+	ImageNet	74.08%	91.67%		3595.13(bs32)		bs x 3 x 224 x 224
100155	SimCLR	CIFAR-10	65.55%			28070(bs32)		bs x 3 x 32 x 32
100156	SkNet50	ImageNet	77.54%			2416(bs8)		bs x 3 x 32 x 32
100157	SMLP	ImageNet	81.25%			298.7(bs8)		bs x 3 x 224 x 224
100158	SPACH	ImageNet	81.5%			462.96(bs8)		bs x 3 x 224 x 224
100159	SpnasNet100	ImageNet	74.19%	91.95%		8408(bs8)		bs x 3 x 224 x 224
100062	SwinTransformer	ImageNet	86.4%	98%		132(bs8)		bs x 3 x 384 x 384
100160	SwinTransformer-tiny	ImageNet	81.15%	95.42%		564.7(bs8)		bs x 3 x 224 x 224
100161	T2T-ViT	ImageNet	81.4%			194.66(bs8)		bs x 3 x 224 x 224
100162	TimeSformer	kinetics400	77.68%			7.53(bs1)		1 x 3 x 3 x 8 x 224 x 224
100163	TNT	ImageNet	81.5%			274(bs8)		bs x 196 x 16 x 24
100164	TResNet	ImageNet		94.43%		3249(bs16)		bs x 3 x 224 x 224
100165	Twins-PCPVT-S	ImageNet	81.22%			613(bs16)		bs x 3 x 224 x 224
100166	Twins-SVT-L	ImageNet	83.7%			175.2209(bs8)		bs x 3 x 224 x 224
100167	VAN	ImageNet	82.78%			874(bs8)		bs x 3 x 224 x 224
100067	VGG19	ImageNet	71.76%	90.80%		1153(bs64)		bs x 3 x 224 x 224
100168	Video-SwinTransformer	kinetics400	80.6%	94.5%		0.607(bs1)		1 x 12 x 3 x 32 x 224 x 224
100169	ViT-small	ImageNet	81.37%			1013(bs8)		bs x 3 x 224 x 224
100170	VOLO	ImageNet	82.53%			124(bs8)		bs x 3 x 224 x 224
100171	VoVNet39	ImageNet	76.77%	93.43%		1767(bs4)		bs x 3 x 224 x 224
100172	Wide-ResNet101	ImageNet	78.86%	94.29%		1151(bs16)		bs x 3 x 224 x 224
100173	Wide-ResNet50	ImageNet	78.48%	94.09%		2097(bs32)		bs x 3 x 224 x 224
100174	Xception	ImageNet	78.8%	94.2%			813(bs8)	bs x 3 x 299 x 299
100175	XCIT	ImageNet	81.86%			443(bs8)		bs x 3 x 224 x 224

CV-detection

ID	Name	Dataset	精度							最优性能(对应bs)		输入shape
ID	Name	Dataset	AP	mAP	DSC-score	F1-score	Top1Acc	ODS	loss	300I Pro		输入shape
100176	3D-UNet	Brats2018			25.6%						10.75(bs1)	bs x 4 x 64 x 64 x 64
100177	AdvancedEAST	天池ICPR				52.08%				137(bs1)		bs x 3 x 736 x 736
100178	AlphaPose	coco		71.47%						1772(bs16)		bs x 3 x 256 x 192
100179	BSN	Activity1.3					74.34%			34617(bs16)		TEM[bs, 400, 100] PEM[bs, 3, 100]
100180	Cascade-MaskRcnn-SwinS	coco		51.4%						3.17(bs1)		1 x 3 x 800 x 1216
100181	CascadeRCNN-DCN101	coco		45%						3(bs1)		1 x 3 x 1216 x 1216
100182	CascadeRCNN-R101-FPN	coco		41.9%						9(bs1)		1 x 3 x 1216 x 1216
100009	CascadeRCNN-ResNet101-FPN-DCN	coco		45%						3.8(bs1)		1 x 3 x 1216 x 1216
100183	CascadeRCNN-ResNet50-FPN	coco		40.5%						6.5(bs1)		1 x 3 x 1216 x 1216
100184	CH-PPOCR-serverV2.0-det	PaddleOCR								154(bs1)		多尺度
100185	CH-PPOCRV2-det	PaddleOCR								232(bs1)		多尺度
100186	CH-PPOCRV3-det	PaddleOCR								215(bs1)		1 x 3 x -1 x -1
100187	CRAFT	随机数								132(bs1)		1 x 3 x 640 x 640
100188	CTPN	ICDAR2013	86.84%							169(bs64)		多尺度
100189	DeepMAR	PETA					78.9%			1642(bs1)		bs x 3 x 224 x 224
100190	EAST-MobileNetV3	ICDAR2015	78.29%							458(bs1)		bs x 3 x 704 x 1280
100191	EAST-ResNet50-vd	ICDAR2015	88.63%							91(bs1)		bs x 3 x 704 x 1280
100192	EfficientDet-D7	coco		53%						6.18(bs1)		bs x 3 x 1536x 1536
100193	EN-PPOCRV3-det	PaddleOCR								168(bs1)		多尺度
100194	FairMOT	MOT17	83.7%							12(bs32)		bs x 3 x 608 x 1088
100029	FasterRCNN-DCN-Res101	coco		44.2%						2.47(bs1)		1 x 3 x 1216 x 1216
100195	FasterRCNN-DCN-Res50	coco		41.1%						8(bs1)		1 x 3 x 1216 x 1216
100030	FasterRCNN-ResNet50	coco		37.2%						14.84(bs1)		1 x 3 x 1216 x 1216
100196	FCENet	icdar2015					87.2%			28.9(bs1)		bs x 3 x 1280 x 2272
100197	Fcos	coco		35.9%						65(bs4)		bs x 3 x 800 x 1333
100198	FOTS	ICDAR2015	86.4%							66(bs16)		bs x 3 x1248 x 2240
100199	FSAF	coco		37.1%						20(bs1)		bs x 3 x 800 x 1216
100200	GFocalV2	coco	40.6%							41.8(bs4)		bs x 3 x 800 x 1216
100201	M2Det	coco					37.8%			65(bs4)		bs x 3 x 512 x 512
100202	NAS-FPN	coco		40.4%						72(bs1)		1 x 3 x 640 x 640
100203	NasNetlarge	ImageNet					82.5%			175(bs4)		bs x 3 x 331x 331
100204	Pelee	VOC										bs x 3 x 304 x 304
100205	PSE-MobileNetV3	ICDAR2015	82.14%							219(bs1)		bs x 3 x 736 x 1312
100206	PSENet-ResNet50-vd	ICDAR2015	85.72%							51(bs1)		bs x 3 x 736 x 1312
100207	Pyramidbox	widerface	95%							8.9(bs1)		1 x 3 x 1000 x 1000
100208	RCF	BSDS500						79.8%			93(bs1)	bs x 3 x 321 x 481 bs x 3 x 481 x 321
100209	RefineDet	VOC2007					79.6%			445(bs16)		bs x 3 x 320 x 320
100210	RetinaMask	coco		27.9%						4.3(bs1)		1 x 3 x 1344 X 1244
100211	RetinaNet	coco		38.3%						17(bs1)		1 x 3 x 224 x 224
100212	RFCN	VOCtest		69.93%						16.52(bs1)		1 x 3 x 1344 X 1344
100213	SFA3D	KITTI							0.603	426(bs4)		bs x 3 x 608 X 608
100214	SSD-MobileNetV1	VOC2007	69.3%							3176(bs4)		bs x 3 x 300 x 300
100215	SSD-MobileNetV2	VOC2007	69.8%							2923(bs4)		bs x 3 x 300 x 300
100218	TextSnake	TextSnake				59%				180.36(bs1)		1 x 3 x 512 x 512
100219	TOOD	coco	42.2%							14.6(bs1)		1 x 3 x 1216 x 1216
100220	VGG16-SSD	VOC					77.26%			751(bs16)		bs x 3 x 300 x 300
100221	YOLOF	coco		42.8%						126(bs16)		bs x 3 x 608 x 608
100222	YOLOR	coco	52.1%							40.9(bs8)		bs x 3 x 1344 x 1344
100223	YOLOX-Tiny	coco		33.1%						684(bs4)		bs x 3 x 640 x 640
100224	YOLOXs	coco		40.1%						890(bs4)		bs x 3 x 640 x 640
100225	YOLOX-MMdetection	coco	51%							77(bs64)		bs x 3 x 640 x 640

CV-segmentation

ID	Name	Dataset	精度							最优性能（对应bs）		输入shape
ID	Name	Dataset	Acc	Liver 1_Dice	AP	mAP	mIOU	maxF	MAE	300I Pro		输入shape
100226	3D-HRNet	Cityscapes					80.83%			9(bs1)		bs x 3 x 1024 x 2048
100227	3D-Nested-UNet	Task03_Liver		96.5%						3.98(bs1)		1 x 3 x 224 x 224
100228	Cascade-MaskRCNN	coco			36.29%					18(bs1)		1 x 3 x 1344 x 1344
100229	Cascade-MaskRCNN-UniFormer	coco				72%				3(bs1)		1 x 3 x 800 x 1216
100230	CascadeRCNN	coco				44.2%				9(bs1)		1 x 3 x 1344 x 1244
100015	DeeplabV3+	VOCtrainval					78.43%			165(bs1)		bs x 3 x 513 x 513
100231	ENet	Cityscapes					54.11%			1327(bs4)		bs x 3 x 480x 480
100232	ErfNet	Cityscapes					72.2%			381(bs8)		bs x 3 x 512 x 1024
100233	FastSCNN	Cityscapes					68.6%			39(bs1)		bs x 3 x 1024 x 2048
100234	FCN-8s	VOC2012					69.01%			84(bs1)		1 x 3 x 500 x 500
100235	GCNet	coco			61%					13(bs1)		1 x 3 x 800 x 1216
100236	ICNet	Cityscapes					68.9%			32(bs8)		bs x 3 x 1024 x 2048
100237	IntraDA	Cityscapes					47.01%			47(bs1)		bs x 3 x 512 x 1024
100238	LPRNet	代码仓提供	90.2%							27313(bs32)		bs x 3 x 24 x 94
100239	MaskRcnn-MMdet	coco				59%				11(bs1)		1 x 3 x 1216 x 1216
100240	Nested-UNet	coco					83.8%			2623(bs4)		bs x 3 x 96 x 96
100241	OCRNet	Cityscapes					79.63%			13(bs1)		bs x 3 x 1024 x 1024
100242	PointRend	Cityscapes					78.85%			1.27(bs1)		1 x 3 x 1024 x 2048
100243	PraNet	kvasir					83.6%			425(bs4)		bs x 3 x 352 x 352
100244	PSPNet	VOC2012					76.18%			67.8(bs16)		bs x 3 x 500x 500
100245	RefineNet	VOC2012					78.6%			87(bs1)		bs x 3 x 500 x 500
100246	Segformer	Cityscapes					75.94%			10.65(bs4)		bs x 3 x 1024 x 2048
100247	Segmenter	Cityscapes					78.89%			3.4(bs1)		bs x 3 x 768 x 768
100248	SeMask	Cityscapes					76.54%			4.7(bs1)		bs x 3 x 1024 x 2048
100249	SETR	Cityscapes					77.35%			3.4(bs1)		1 x 3 x 768 x 768
100250	SiamMask	VOT2016					42.7%				302(bs1)	多尺度
100251	SOLOV1	coco			32.1%					10(bs1)		1 x 3 x 800 x 1216
100252	STDC	Cityscapes					71.81%			27(bs1)		1 x 3 x 1024 x 2048
100253	Swin97	ADE20K					44.78%			21(bs1)		bs x 3 x 512 x 512
100254	Swin98	ADE20K					47.92%			16(bs1)		bs x 3 x 512 x 512
100255	Swin99	ADE20K					48.29%			14(bs1)		bs x 3 x 512 x 512
100256	Swin100	ADE20K					48.74%			14(bs1)		bs x 3 x 512 x 512
100257	SwinTransformer-Semantic-Segmentation	ADE20K					48.06%			19(bs1)		1 x 3 x 512 x 512
100258	Transformer-SSL	coco			68.8%					4(bs1)		1 x 3 x 800 x 1216
100066	U2Net	ECSSD						94.8%	0.033	240(bs1)		bs x 3 x 320 x 320
100259	Ultra-Fast-Lane-Detection	Tusimple	95.8%							2254(bs32)		bs x 3 x 288 x 800
100260	VNet	LUNA16	99.4%							44(bs1)		bs x 64 x 80 x 80
100261	Wseg	VOC					62.7%			5(bs1)		bs x 3 x 1020 x 1020
100262	YOLACT	coco				32.07%				163(bs1)		bs x 3 x 550 x 550
100263	YOLACTEdge	coco				27.96%				270(bs1)		bs x 3 x 550 x 550
100264	YOLACT++	coco				34.9%					31(bs8)	bs x 3 x 550 x 550

CV-face

ID	Name	Dataset	精度		300I Pro最优性能（对应bs）	输入shape
ID	Name	Dataset	Acc	mAP	300I Pro最优性能（对应bs）	输入shape
100265	AlignedReID	Market1501	80.55%		5293(bs32)	bs x 3 x 256 x 128
100266	Centroids-ReID	DukeMTMC-reID	96.8%		4287(bs8)	bs x 3 x 256 x 128
100267	FaceBoxes	FDDB	94.8%		2332(bs1)	bs x 3 x 224 x 224
100268	FaceNet	LFW	99.2%		7964(bs16)	bs x 3 x 160 x 160
100269	ReID-PCB-baseline	Market	92.1%		2031(bs16)	bs x 3 x 384 x 128
100270	ReId-MGN	Market		94.23%	1519(bs8)	bs x 3 x 384 x 128
100271	Retinaface	WiderFace	87.56%		1502(bs16)	bs x 3 x 1000 x 1000

CV-gan

ID	Name	Dataset	精度			300I Pro最优性能（对应bs）	输入shape
ID	Name	Dataset	IS	FID	Acc	300I Pro最优性能（对应bs）	输入shape
100272	BigGAN	噪声数据	94.009	10		544(bs16)	bs x 1 x 20 bs x 5 x 148
100273	CGAN	随机数				1935(bs1)	1 x 100 x 72
100274	CycleGAN	maps			CycleGAN_Ga 1 CycleGAN_Gb 0.99	CycleGAN_Ga 231(bs64) CycleGAN_Gb 232(bs64)	bs x 3 x 256 x 256
100275	DCGAN	噪声数据			1	108781(bs32)	bs x 100 x 1 x 1
100276	DGNet	Market-1501		18.12		584(bs8)	bs x 1 x 256 x 128 bs x 3 x 256 x 128
100277	GAN	随机数				496239(bs64)	bs x 100
100278	Pix2Pix	facades				963(bs32)	bs x 3 x 256 x 256
100279	Pix2PixHD	cityscapes				5(bs1)	bs x 36 x 1024 x 2048

CV-image_process

ID	Name	Dataset	精度	300I Pro最优性能（对应bs）	输入shape
ID	Name	Dataset	PSNR	300I Pro最优性能（对应bs）	输入shape
100280	CrossScale-NonLocal-Attention	Set5	32.57	0.71(bs1)	1 x 3 x 56 x 56
100281	DnCNN	dncnn	31.53	166(bs16)	bs x 1 x 481 x 481
100282	SRFlow	DIV2K	23	0.7(bs1)	1 x 3 x 256 x 256
100283	WDSR	DIV2K	34.75	13(bs1)	bs x 3 x 1020 x 1020

CV-image_registration

ID	Name	Dataset	精度	300I Pro最优性能（对应bs）	输入shape
ID	Name	Dataset	Acc	300I Pro最优性能（对应bs）	输入shape
100284	SuperPoint	HPatches	80.6%	2528(bs8)	bs x 1 x 240 x 320

CV-image_retrieval

ID	Name	Dataset	精度	300I Pro最优性能（对应bs）	输入shape
ID	Name	Dataset	mAP	300I Pro最优性能（对应bs）	输入shape
100285	BLIP	coco	81.3%	text:1662(bs64) image:72(bs1) image_feat:73(bs1)	text:bs x 35 image:bs x 3 x 384 x 384 image_feat:bs x 3 x 384 x 384

CV-pose_estimation

ID	Name	Dataset	精度				300I Pro最优性能（对应bs）	输入shape
ID	Name	Dataset	Top1Acc	Top5Acc	AP	MPJPE	300I Pro最优性能（对应bs）	输入shape
100286	3DMPPE-RootNet	MuPoTS	31.81%				1565(bs4)	bs x 3 x 224 x 224 bs x 1
100287	DEKR	coco			67.7%		7.72(bs1)	多尺度
100288	HigherHRNet	coco			67.1%		185(bs1)	多尺度
100289	HRNet-MMLab	coco			65.3%		151(bs1)	多尺度
100290	MSPN	coco			74.1%		933(bs4)	bs x 3 x 256 x 192
100291	PoseC3D	HMDB51	69.22%	91.31%			22.3(bs8)	bs x 20 x 17 x 48 x 56 x 56
100292	TransPose	coco			73.7%		500(bs4)	bs x 3 x 256 x 192
100293	UniFormer	coco			93.5%		295(bs8)	bs x 3 x 256 x 192
100294	VideoPose3D	Human3.6M				46.6	280257(bs2)	2 x 6115 x 17 x 2

CV-quality_enhancement

ID	Name	Dataset	精度			300I Pro最优性能（对应bs）	输入shape
ID	Name	Dataset	ACC	PSNR	SSIM	300I Pro最优性能（对应bs）	输入shape
100295	ADNet	BSD68	29.24%			215(bs64)	bs x 1 x 321 x 481
100296	SRGAN	Set5		33.4391	93.08%	380(bs8)	bs x 3 x 140 x 140

CV-super_resolution

ID	Name	Dataset	精度		300I Pro最优性能（对应bs）	输入shape
ID	Name	Dataset	PSNR	SSIM	300I Pro最优性能（对应bs）	输入shape
100297	FLAVR	UCF101	29.83	94.46%	77(bs16)	bs x 3 x 224 x 224
100298	RCAN	Set5	38.25	96.06%	12(bs1)	bs x 3 x 256 x 256
100299	RDN	Set5	38.27		47(bs1)	bs x 3 x 114 x 114
100300	Real-ESRGAN	代码仓提供			251(bs4)	bs x 3 x 220 x 220 bs x 3 x 64 x 64
100301	SRCNN	Set5	36.33		2361(bs1)	bs x 1 x 256 x 256

CV-tracking

ID	Name	Dataset	精度					最优性能（对应bs）		输入shape
ID	Name	Dataset	success_score	precision_score	Acc	EPE	MAPE	300I Pro		输入shape
100302	FlowNet2	MPI-Sintel-complete				2.15		14(bs1)		bs x 3 x 448 x 1024
100303	SiamFC	OTB2015	57.2%	76.2%				exemplar_bs1:6072(bs1) search_bs1:948(bs1)		1 x 3 x 255 x 255 1 x 9 x 127 x 127
100054	SiamRPN	VOT			63.9%				42(bs1)	1 x 3 x 127 x 127 1 x 3 x 255 x 255
100304	GMA	MPI-Sintel-complete			final:88.95% clean:92.65%			0.77(bs4)		1 x 3 x 440 x 1024
100305	DeepCTR	代码仓提供					WDL:2.147 xDeepFM:1.97 AutoInt:2.14	WDL:0.079(bs40) xDeepFM:0.177(bs40) AutoInt:0.22(bs40)		40 x 6

CV-video_understanding

ID	Name	Dataset	精度				300I Pro最优性能（对应bs）	输入shape
ID	Name	Dataset	ADK	MKR	Top1Acc	Top5Acc	300I Pro最优性能（对应bs）	输入shape
100306	FOMM	taichi	6.7975	0.036			kp detector:957(bs1) generator:7(bs1)	kp detecto:1 x 3 x 256 x 256 generator:多尺度
100307	I3D-Nonlocal	kinetics400			70.03%	89.51%	14.39(bs1)	bs x 10 x 3 x 32 x 256 x 256
100308	NonLocal	kinetics400			71.62%	90.27%	97(bs1)	bs x 3 x 8 x 224 x 224
100309	SlowFast	kinetics400			70.07%	88.55%	138(bs1)	bs x 3 x 32 x 224 x 224
100310	TSM	UCF-101			94.48%	99.63%	194(bs1)	bs x 8 x 3 x 224 x 224
100311	TSN	UCF-101			82.83%		22.19(bs32)	bs x 75 x 3 x 256 x 256
100312	X3D	Kinetic400			73.75%	90.25%	386(bs8)	bs x 3 x 13 x 182 x 182

Audio

ID	Name	Dataset	精度				最优性能（对应bs）		输入shape
ID	Name	Dataset	EER	WER	ROC_AUC	mel_loss	300I Pro		输入shape
100313	AASIST-L	LA	0.979				168(bs64)		1 x 64600
100314	Data2Vec	LibriSpeech		0.94			11(bs1)		bs x 559280
100021	Ecapa-TDNN	VoxCeleb1			0.9991		1654(bs4)		bs x 80 x 200
100031	FastPitch	LJSpeech				11.33	126(bs8)		bs x 200
100315	Jasper	LibriSpeech		9.709			41(bs1)		bs x 64 x 4000
100316	LSTM	timit		18.9075			83.4(bs64)		bs x 390 x 243
100317	RawNet2	VoxCeleb1	2.5%					77(bs16)	bs x 59049
100318	Speech-Transformer	aishell	9.9%					0.82(bs1)	多尺度
100319	TDNN-contrib	librispeech		98.69%				1562(bs16)	多尺度
100071	Wav2Vec2	librispeech		2.96%			157(bs16)		bs x 10000
100320	WaveGlow	LJSpeech					3(1 x 80 x 154)		多尺度
100321	WeNet	aishell		4.68%			154.8(bs1)		多尺度
100321	whisper	librispeech_asr_dummy		8.21%				67.32(bs1)	bs x 80 x 3000

Knowledge

ID	Name	Dataset	精度	300I Pro最优性能（对应bs）	输入shape
ID	Name	Dataset	MRR	300I Pro最优性能（对应bs）	输入shape
100322	RotatE	FB15k-237	0.3355	head:222(bs64) tail:222(bs64)	bs x 3 bs x 14541

Embedding

ID	Name	最优性能（对应bs）		输入shape
ID	Name	300I DUO	800I A2	输入shape
100410	bge-m3	forward 23.23ms, e2e 137.59ms (bs2)	forward 14.71ms, e2e 103.88ms (bs2)	动态输入
100411	bge-reranker-v2-m3	forward 22.57ms, e2e 60.47ms (bs2)	forward 15.08ms, e2e 46.64ms (bs2)	动态输入

Nlp

ID	Name	Dataset	精度						300I Pro最优性能（对应bs）	输入shape
ID	Name	Dataset	Acc	WER	loss	BLEU	F1	bpc	300I Pro最优性能（对应bs）	输入shape
100005	BertBase-Uncased	squad	88.78%						221(bs4)	bs x 512
100323	BertSum	代码仓提供	42.85%						138.09(bs8)	bs x 512 bs x 37
100324	CH-PPOCR-serverV2.0-rec	PaddleOCR							289(bs1)	多尺度
100325	CH-PPOCRV2-rec	PaddleOCR							260(bs1)	多尺度
100326	CH-PPOCRV3-rec	PaddleOCR							1411(bs1)	多尺度
100327	CNN-Transformer	Librispeech		0.0556					52(bs1)	多尺度
100328	DeBERTa	MNLI	90.46%						5(bs64)	bs x 256
100329	ELMO	1 Billion Word							62(bs1)	1 x 8 x 50
100330	EN-PPOCRV3-rec	PaddleOCR							6144(bs16)	bs x 3 x 48 x 320
100331	GPT2	wiki_zh_2019			16.5				189(bs16)	bs x 512
100332	GNMT	newstest2014				22.69			24(bs1)	1 x 1 1 x 30
100333	HuBERT	test-clean		2.13					3(bs1)	1 x 580000
100334	RARE-ResNet34-vd	LMDB	84.79%						1603(bs32)	bs x 3 x 32 x 100
100335	Rosetta-MobileNetV3	LMDB	77.38%						24219(bs64)	bs x 3 x 32 x 100
100336	Rosetta-ResNet34-vd	LMDB	80.63%						7418(bs16)	bs x 3 x 32 x 100
100337	SATRN	IIIT5K	94.87%						31(bs1)	1 x 3 x 32 x 100
100338	SpanBERT	SQuAD 1.1					93.95%		43(bs1)	bs x 512
100339	StarNet-MobileNetV3	LMDB	80.02%						2617(bs64)	bs x 3 x 32 x 100
100341	TinyBERT	SST-2	92.32%						11160(bs64)	bs x 4 x 84 x 84
100342	Transformer	Multi30k	40.92%						48(bs1)	1 x 15
100343	TransformerXL	enwik8						1.966	287(bs1)	80 x 1 160 x 1 x 512
100344	TrOCR	IAM		4.25					8(bs1)	1 x 3 x 384 x 384
100345	VilBERT	coco					0.67		493(bs32)	多尺度

ID	Name	Dataset	精度	300I Pro最优性能（对应bs）	输入shape
ID	Name	Dataset	Acc	300I Pro最优性能（对应bs）	输入shape
100346	C51	随机数	98.9%	6050(bs1)	1 x 4 x 84 x 84
100347	DQN	随机数	100%	8147(bs1)	1 x 4 x 84 x 84

Ascend/ModelZoo-PyTorch

欢迎使用Ascend ACL_PyTorch

如何贡献

支持模型列表（按字母顺序排序）

规范模型

生态贡献模型

简介

发行版

贡献者

语言

近期动态

Ascend/ModelZoo-PyTorch .gitee-modal { width: 500px !important; }

欢迎使用Ascend ACL_PyTorch

如何贡献

支持模型列表（按字母顺序排序）

规范模型

生态贡献模型

简介

发行版

贡献者

语言

近期动态

搜索帮助

Ascend/ModelZoo-PyTorch