It was thought of by the open source artificial intelligence (AI) project of a well-known domestic enterprise
1. Well-known open source AI projects of domestic large and medium-sized companies (as far as I know, small and medium-sized companies with more than 100 stars are counted)
| company | Number of well-known projects | AI-related accounts | well-known projects |
|---|---|---|---|
| Baidu | 21 | Baidu、ApolloAuto、PaddlePaddle | AI framework (Paddle/Lite/Serving), self-driving Apollo, Paddle application series (OCR/NLP/Detection/Speech/Seg/Clas/Rec/Video/GAN/Hub), NLP series (ERNIE/LAC/AnyQ/Senta/ DDParser/PLATO), reinforcement learning (PARL) |
| Huawei | 15 | Huawei, huawei-noah, mindspore-ai | AI framework (mindspore/serving), mind series (armour/quantum/science/Hub), CV field (GhostNet/Efficient-AI-Backbones/vision), NLP field (Nezha/TinyBERT/PanGu-α/Noah_WuKong/DynaBERT/PMLM ), graph neural network (graphengine), reinforcement learning RL (SMARTS) |
| Tencent | 13 | Tencent、tencent-ailab | AI framework (ncnn reasoning/TNN lightweight reasoning/TurboTransformers reasoning), CV field (FeatherCNN/MedicalNet/ObjectDetection-OneStageDet/FaceDetection-DSFD/tencent-ml-images), NLP field (Tencent-WordEmbedding/NeuralNLP-NeuralClassifier/TexSmart) , audio domain (bddm), reinforcement learning RL (PhoenixGo) |
| Ali Baba | 7 | Alibaba、Alibaba-NLP、Alipay | AI framework (MNN reasoning), EASY series - recent (CV/NLP/REC/RL), AliceMind (StructBERT, LatticeBERT) |
| JD | 5 | JDAI、JDAI-CV | AI framework (dabnn reasoning), CV field (fast-reid/FaceX-Zoo), nlp_baai (JDAI-BERT/JDAI-Word-Embedding) |
| XiaoMi | 3 | Xiaomi | AI framework (mace lightweight reasoning), MiNLP, kaldi-onnx |
| ByteDance | 2 | Bytedance | AI framework (byteps distributed training/lightseq sequence task training) |
| DiDi | 2 | DiDi、Delta-ML | AI framework (dlflow distributed training), reading comprehension delta |
| Meituan | 1 | Meituan | CV field (YOLOv6) |
| NetEase | 1 | netease-youdao, NetEase-GameAI | Face2FaceRHO、EMLL |
| Trip | 1 | ctripcorp | C-OCR |
| iFLYTEK | 1 | iFlytekJudiciary | NLP field (Iflytek long text classification dataset/CAIL2019_CJRC) |
| 360 | 0 | Qihoo360 | TensorNet |
| Kingsoft | 0 | kingsoft-wps | KSAI-Lite |
| PDD | None | / | / |
| 58 | None | / | / |
| Theirs | None | / | / |
| Swamp | None | / | / |
| SenseTime | 14 | OpenMMLab、Sense-GVT、SenseTime X-Lab | mm系列(mmcv/mmdetection/mmediting/mmsegmentation/mmocr/mmdetection3d/mmselfsup/mmaction/mmclassification/mmgeneration/mmdeploy)、DeCLIP、X-Temporal、CV领域(Deformable-DETR) |
| for example | 10 | MegEngine、Megvii-BaseDetection、megvii-model | AI framework (MegEngine), CV field (YOLOX/YOLOF/ShuffleNet series/IOU-loss/BorderDet/DeFCN/LearningToPaint/ML-GCN/BBN/neural-painter) |
| that is | 1 | YITUTech | T2T-ViT |
| cloudwalk | 0 | CloudWalk | ddbscan |
| zhuyi | 6 | ZhuiyiTechnology | NLP field (nl2sql_baseline/simbert/WoBERT/roformer/GlobalPointer), pretrained model pretrained-models |
| shannonai | 4 | Shannon AI | NLP领域(ChineseBert/mrc-for-flat-nested-ner/glyce/dice_loss_for_NLP) |
| chumenwenwen | 1 | wenet-e2e | Audio field (wenet/speech-synthesis-paper/wekws/opencpop) |
| speech | 1 | aispeech-lab | Audio field (advr-avss/VisBCI/TinyWASE) |
| memect | 1 | memect | Knowledge map field (kg-beijing) |
| leyantech | 0 | leyantech | / |
| datagrand | None | / | / |
| emotibot | None | / | / |
| hivoice | None | / | / |
| tigerobo | None | / | / |
Two, feeling
或许统计不全面，不过依然可以看出：
1、国内开源尤其是AI开源确实不太活跃，相对于企业，更多的可能是大学和个人。
2、百度的Ai开源项目最多，确实反映了百度AI在国内互联网公司中最强的实力。
3、相较于创业公司，大公司开源动力更强，不过与美国头部互联网公司相比，国内公司的实力确实是不强，有影响力的项目更是少之又少。
4、不过与大众印象中阿里国内最强的开源互联网企业不同，阿里在AI领域的开源并不活跃，在CV、NLP似乎也见不到阿里的经典项目和论文。
5、中小型互联网公司在AI领域的开源动力也不强，开源项目是少之又少。
6、反倒是一些初创企业开源意愿较强，也是宣传和证明自己技术实力的一种方式吧，例如CV领域的AI四小龙中的商汤、旷视，又比如NLP领域的追一科技和香侬科技。