how to recognize peoples face or body or the close-up for specify portion of body batchly
how to recognize peoples face or body or the close-up for specify portion of body batchly
As my site is placed on the domestic server, the last time was blocked is 24 years in May, I do not know is peer reporting, or tencent cloud has a monitoring mechanism, will scan the content of your site, and then distributed to the manual review, finally my site in May, was tencent to rectify the reason for the downgrade, I was very angry, but the follow-up due to the out of the engage in a lot of offline discipleship courses, but also subject to the then have not thought of a better compromise model, so the site has never been online, all the kubernetes course recording and some of the upcoming online features of the development, are also in the local virtual machine cluster environment, until this time I put the site back online.
This site, no matter what part, in fact, a large number of applications of artificial intelligence (for example, a lot of video generation, completion is dependent on me based on so-vits-svc training out of my own voice, but also based on sd-wav2lop-uhq to do a short video to a long video based on the face of the generation of the lip-sync), such as data collection, which is a lot of articles from a number of foreign senior emotional blog with rod type of golang crawler did data collection, given colly only for static url address is effective, and many foreign sites are in fact their own written framework, or with drupal this heavy artillery, or wordpress plus a lot of anti-climbing technology, so with the conventional way (scrapy of python. puppeteer is also good, but heavy writing javascript, playwright selenium are python system, now unless writing artificial intelligence gadgets related) is difficult to deal with these sites with complex interactions, so I used the golang system of some of the frameworks to engage in these crawlers, the main focus on the concurrency of high performance, and there can be a lot of hack operations, and the article extracted back to the ordinary way of translation on the language is too poor, surely to manually intervene, I am still thinking more chatgpt way to do this thing, but in short, ordinary site behind, in fact, the minutiae, are manifested with the times with the walk of the remarkable features. Back to the topic, the back of the idea of this way, you do not put pictures, the article readability is low, at least I feel so!
You think about it, you small program to open an article, through a picture does not, boring! But put a picture of it, I do not know why my self-media transport feel not too good, because I oil pipe has been blocked five times, the same is powerful to do hitched dating girls have to eat operation these, I do not know why was blocked always me! Harm, it’s a long story, I was very early, using python to generate more than 2,000 videos, instantly over the domestic watermelon platform revenue threshold, then also wrote an article to celebrate, move here:23,000 traffic in 4 days successfully opened in the video plan video editing to earn money not shabby attached puppeteer technology list to help fully automated video editingI was so disheartened that I switched to tubes, and my first account quickly passed 2000 subscriptions and airtime, but soon the channel was removed as a whole, and it was complicated, and I had the opportunity to share the cause and effect of my several channel blockings in video form! I’ve had 4 or 5 channels removed, it’s really heartbreaking! Crooked, back to the topic, so I applied Gaussian blur to the image as a whole, you can tell there’s an image here anyway, but then there’s no actual panoramic display, it shouldn’t be easy to get down to brass tacks either, so that’s what I said I’d do!
In response to the picture or video mosaic, now more formed program is to add a mask layer, belonging to the simple part, the complex part lies in the face recognition. But yovov3, yovov5 came out of nowhere, this is no longer a problem, I have searched for foreign articles, in fact, yovov5 do face recognition, should have reached the peak of the iterative version of the latter, the solution is no longer gathered in the face recognition, so I first use yovov5 to do the test, but the effect can be said to have no, may be some of the reasons for the parameters of the configuration, in short, I did not detect very well. Not very good detection, I probably tens of thousands of pictures put in to do the test, almost all can not correctly recognize the face, may also need to read the document carefully again! As my local cuda is 11.8 version, in fact, now nvidia now cuda version has 12 +, so I feel that it may be a version compatibility issues, so I read the yovov3 document, intend to use this try to see if there can be progress, do not try not know, a try to shock, the effect is very good, but for the extremely tricky angle of the side of the face, there are still Leakage phenomenon, but the problem is not big, at the same time I strengthened its function, after all, the picture in my article, it seems that the body part should also do masking, after all, really afraid of being marked as yellow, it is very annoying!
Finally put a few pictures, you feel it!
源码我会放在我的知识星球！
