dongxiaolong comments

Results 22 comments of


                                            dongxiaolong

ImportError: No module named 'openvino'

I'm also met the problem. Have you already solve the problem?

ImportError: No module named 'openvino'

I may already find the bugs, on the 2019 R1 version ,the openvino module is included by the armv7l, should import openvino by from armv7l.openvino.inference_engine import IENetwork, IEPlugin

ImportError: cannot import name 'roi_align_cuda'

I alse get the same trouble.Have you solved it ?

kp-bbox rescale problem

![crowdpose_100024_kapao_l_coco_kpbbox_kp_obj](https://user-images.githubusercontent.com/23193031/168944078-ff2b81e1-b9b4-424a-9473-869bae4f81c6.png)

kp-bbox rescale problem

![crowdpose_100024_kapao_l_coco_bbox_pose_kpbbox_kp_obj](https://user-images.githubusercontent.com/23193031/168944186-a25c6405-db1e-42b0-8710-81494cc2f99c.png)

kp-bbox rescale problem

> I think it rescales kp-bboxs 2 times. At line 88 in demos/image.py, function post_process_batch(...) was called. In this function, the mutable object kp_dets do the rescale calculation and some...

支持中英文双语的模型？

> 另外，方便询问一下您的使用场景吗？如果是普遍需求的话，我们也会考虑下一版模型同时支持中英文双语。比如现在很多的论文阅读场景都会有需求

> 另外，方便询问一下您的使用场景吗？如果是普遍需求的话，我们也会考虑下一版模型同时支持中英文双语。你好，我觉得现在这类模型的需求，应该是支持多语言长文本相似度召回的，目前开源模型的Bert base model 的sequence length 都只有512，而OpenAI的ada2是8196的，多语言能力和sequence length 的差距巨大。综合这两个方面的原因，目前开源模型只在单语言短文本相似度的对比上有优势，使用场景是十分受限的。

[Bug]: Ray memory leak

The same situation occurs during multi-GPU deployment.

Text Embeddings Inference is now Open Source!

Congrats! This will get more people to use it!