模型推理
更新时间：2024-05-17
模型推理
模型支持列表
目前DK-1B主要支持飞桨系列的PaddleClas、PaddleDetection、PaddleSeg、PaddleOCR套件。其中，有支持NPU推理加速的模型和支持CPU通用推理的模型，详单如下：
NPU模型支持列表
| 模型类型 | 模型列表 | 支持尺寸 | 模型信息 |
|---|---|---|---|
| clas_resnet | ResNet18 ResNet34 ResNet50 |
224x224 | PaddleClas |
| det_yolov3 | YOLOv3-Mobilenet-V1 YOLOv3-DarkNet53 YOLOv3-ResNet34 |
320x320 416x416 608x608 |
PaddleDetection |
| kpt_hrnet | HRNet-w32 | 256x192 384x288 |
PaddleDetection |
CPU模型支持列表
| 套件名称 | PaddleClas | PaddleDetection | PaddleSeg | PaddleOCR | ||||
|---|---|---|---|---|---|---|---|---|
| Github链接 | PaddleClas | PaddleDetection | PaddleSeg | PaddleOCR | ||||
| 细分 | classification | Detection | Multi Object Tracking | KeyPoint Detection | Semantic Segmentation | Image Matting | detection | recognition |
| 1 | ShiTuV2 | PP-PicoDet | DeepSORT | PP-TinyPose | Glore | PP-Matting | SAST | SVTR |
| 2 | PP-LCNetV2 | PP-YOLOE+ | HRNet | LRASPP | PP-HumanMatting | ch-PP-OCRv3 | StarNet | |
| 3 | PP-HGNet | PP-YOLOE | Lite-HRNet | DDRNet | MODNet | PSE | Rosetta | |
| 4 | CSWinTransformer | PP-YOLO | PP-LiteSeg | DB | ch_PP-OCRv3 | |||
| 5 | ReID | YOLOv3 | DeepLabV3 | EAST | ch_PP-OCRv2 | |||
| 6 | PULC | FCOS | DeepLabV3P | RARE | ||||
| 7 | PVT_V2 | Faster RCNN | GCNet | CRNN | ||||
| 8 | ResNet | SFNet | SRN | |||||
| 9 | GhostNet | LaneSeg | ||||||
| 10 | Twins | OCRNet | ||||||
| 11 | ResNeSt | DecoupledSegNet | ||||||
| 12 | DPN | PointRend | ||||||
| 13 | ViT | EMANet | ||||||
| 14 | MixNet | U-Net | ||||||
| 15 | DarkNet53 | ESPNetV1 | ||||||
| 16 | SEResNeXt | DANet | ||||||
| 17 | SqueezeNet | SegFormer | ||||||
| 18 | Inception | PFPNNet | ||||||
| 19 | VGG | FCN | ||||||
| 20 | MobileNetV2 | BiSeNetV2 | ||||||
| 21 | ReXNet | ANN | ||||||
| 22 | DenseNet | STDCSeg | ||||||
| 23 | MobileNetV1 | ENCNet | ||||||
| 24 | HarDNet | HarDNet | ||||||
| 25 | Xception | DNLNet | ||||||
| 26 | AlexNet | HRNetW48Contrast | ||||||
| 27 | Res2Net | |||||||
| 28 | ShuffleNetV2 | |||||||
| 29 | DLA | |||||||
| 30 | PP-LCNet | |||||||
| 31 | MobileNetV3 | |||||||
| 32 | DeiT | |||||||
| 33 | EfficientNet | |||||||
| 34 | HRNet | |||||||
| 35 | ESNet | |||||||
| 36 | CSWinTransformer | |||||||