[FFmpeg-devel] =?gb18030?b?u9i4tKO6ICBbUEFUQ0ggdjFdIGxpYmF2Zmkv?= =?gb18030?q?dnn=3A_add_Paddle_Inference_as_one_of_DNN_backend?=

[=?gb18030?B?V2VuemhlV2FuZw==?= <wongwwz@foxmail.com>]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="gb18030", Size: 5578 bytes --]

Thank you for your reply and I am very glad to receive your opinion.

For me,github star is not very important, I just want to use it to highlight how many people are using deep learning frameworks right now. However, I have learned about dnn's plans for other frameworks.

I think it would be a good idea to add a glue layer and hope to see it implemented soon. Do I have the opportunity to participate in the development of this module?

Thank you again for answering my question.


best,
Wenzhe



WenzheWang
wongwwz@foxmail.com



 




------------------ Ô­Ê¼Óʼþ ------------------
·¢¼þÈË:                                                                                                                        "FFmpeg development discussions and patches"                                                                                    <yejun.guo-at-intel.com@ffmpeg.org&gt;;
·¢ËÍʱ¼ä:&nbsp;2023Äê5ÔÂ10ÈÕ(ÐÇÆÚÈý) ÍíÉÏ7:30
ÊÕ¼þÈË:&nbsp;"FFmpeg development discussions and patches"<ffmpeg-devel@ffmpeg.org&gt;;

Ö÷Ìâ:&nbsp;Re: [FFmpeg-devel] [PATCH v1] libavfi/dnn: add Paddle Inference as one of DNN backend





&gt; -----Original Message-----
&gt; From: ffmpeg-devel <ffmpeg-devel-bounces@ffmpeg.org&gt; On Behalf Of
&gt; "zhilizhao(ÕÔÖ¾Á¢)"
&gt; Sent: Wednesday, May 10, 2023 12:09 PM
&gt; To: FFmpeg development discussions and patches <ffmpeg-
&gt; devel@ffmpeg.org&gt;
&gt; Subject: Re: [FFmpeg-devel] [PATCH v1] libavfi/dnn: add Paddle Inference as
&gt; one of DNN backend
&gt; 
&gt; 
&gt; 
&gt; &gt; On May 10, 2023, at 10:25, WenzheWang <wongwwz@foxmail.com&gt; wrote:
&gt; &gt;
&gt; &gt; Dear Madam or Sir,
&gt; &gt;
&gt; &gt;
&gt; &gt; Hope this email finds you well.
&gt; &gt;
&gt; &gt;
&gt; &gt; I am writing this email since i recently found FFmepg remove DNN native
&gt; backend, and i will be really grateful if you let me know if there is&nbsp; any new
&gt; plan on libavfilter/dnn.
&gt; &gt;
&gt; &gt;
&gt; &gt; I would like to explain to you again about the addition of dnn paddle
&gt; backend.
&gt; &gt;
&gt; &gt; At&nbsp; present, ffmpeg only supports openvino and tensorflow backend.
&gt; Among&nbsp; the current deep learning frameworks, TensorFlow is the most active
&gt; in&nbsp; development. TensorFlow has 174k stars and pytorch has 66.5k. openvino
&gt; is 4.2k, and the models that openvino can implement are relatively few.&nbsp; But
&gt; in terms of attention on GitHub, there's no doubt that TensorFlow&nbsp; and
&gt; pytorch are more promising. Currently, the paddle framework has&nbsp; reached
&gt; 20.2k stars on github, which is much more widely used and active&nbsp; than
&gt; frameworks such as mxnet and caffe.
&gt; 
&gt; Stars don't matter much here.
&gt; 
&gt; Just for reference, there is a thread before:
&gt; 
&gt; https://patchwork.ffmpeg.org/project/ffmpeg/patch/20220523092918.9548-
&gt; 2-ting.fu@intel.com/
&gt; 
&gt; &gt;
&gt; &gt; Tensoflow has a very&nbsp; rich ecosystem. The TensorFlow models library
&gt; updates very quickly and&nbsp; has existing examples of deep learning applications
&gt; for image&nbsp; classification, object detection, image generation text, and
&gt; generation&nbsp; of adversus-network models. The dnn libavfilter module is
&gt; undoubtedly very necessary for tensorflow&nbsp; backend to support. But the
&gt; complexity of the TensorFlow API and the&nbsp; complexity of the training are
&gt; almost prohibitive, making it a love-hate&nbsp; framework.
&gt; &gt;
&gt; &gt; PyTorch framework tends to be applied to academic&nbsp; fast implementation,
&gt; and its industrial application performance is not&nbsp; good. For example, Pytorch
&gt; framework makes a model to run on a server,&nbsp; Android phone or embedded
&gt; system, and its performance is poor compared&nbsp; with other deep learning
&gt; frameworks.
&gt; &gt;
&gt; &gt;
&gt; &gt; PaddlePadddle&nbsp; is an open source framework of Baidu, which is also used
&gt; by many people&nbsp; in China. It is very consistent with the usage habits of
&gt; developers,&nbsp; but the practicability of the API still needs to be further
&gt; strengthened. However, Paddle is the only deep learning framework I&nbsp; have
&gt; ever used, which does not configure any third-party libraries and&nbsp; can be
&gt; used directly by cloning make. Besides, Paddle occupies a small&nbsp; amount of
&gt; memory and is fast. It also serves a considerable number of&nbsp; projects inside
&gt; Baidu, which is very strong in industrial application.&nbsp; And PaddlePaddle
&gt; supports multiple machine and multiple card training.

Imo, my idea is that we can add 1 or 2 dnn backends as discussed at 
http://ffmpeg.org/pipermail/ffmpeg-devel/2022-December/304534.html

The background is that we see different good models from different deep learning
frameworks, and most framework does not support models developed with other 
frameworks due to different model format. imo, we'd support several popular frameworks.


_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".