I've been looking to program video editing as it's more fun and teaches a lot more in gamedev than using pipeline and tedious and always crashing gui apps. Thanks for these
It really depends... You have more flexibility like this (editing software can't have a button/setting for absolutely everything). But using an editor is probably faster most of the time.
I have good knowledge in JS, bat i don't know so about canvas and in this video opened very nice idea how to use basic js code for video player. You very cool explain. For example Google not finding result about machine language wrote in vinil js. Second. It is canvas. I finded your channel of random when watch other tematik with programming (Sorry. Maybe wrot with mistakes. I try)
I also like that you don't use other people's frameworks. All other sources point to ready-made libraries with machine intelligence, but no one understands how it works. And it is important for me to understand exactly how the neural network works. For example, I did duplicate music recognition (even if the file is compressed or passed through the equalizer). I did it like this. Got the difference between the volume levels on equal sections of time. If the difference is small, then awarded positive points (+1, +5). The maximum number cannot exceed, for example, 100. If the difference is large, then the penalty (-50, -100 points) was triggered. If the result in the cycle was less than 200, then it is not a duplicate. 80GB of music, the search for a duplicate lasted 1 second. But I still used a very compressed base of my own format. My program also recognized fonts or icons from the screenshot. I made the rules. If it matches at the point I specify, then I reward +1, if it does not match, then I punish -20. Few points. There are also anti-points. These are places where there should be no point. I can confidently say that my method works very quickly and well, but the neural network understands the gradients better and the neural network can be improved to a certain level - this is an advantage. My advantage is a completely controlled process and understanding from beginning to end and at any stage. By the way, RU-vid once offered an algorithm for recognizing duplicate music. Nobody wrote me off, but it may have happened - they did it a month after my letter.
@@Radu I did not find a unique optimization formula on my own. I worked on the principle of rules and layers. The next layer simply had the highest priority. But it worked very quickly. Briefly described the rules, traced whether there were any errors, re-amended the rules and everything is ready. I did not like the neural network. She is good, but not perfect. The neural network is indispensable in many ways, but it lacks logic - it's math and random numbers. Without random numbers, everything will break. For example, a person sits in an excavator and tries different control levers. Try one neatly. Super. It's safe. Second. Also safe. And so the process is set up. The neural network will break 10,000 cars to learn to drive. How can this process be optimized so that there are fewer victims and progress moves upwards? There is no difference between "wrong" and "very wrong" and "smell wrong". And this is a problem. And where is the computer logic with zeros and ones? Where is the logic as in chess with a search of all possible values? No. The neural network is mathematics and with a step of 0.1 or 0.001. It is not necessary to transfer all problems only to a neural network. I think so. There must be another way. Maybe a hybrid. I also did a contour search on the image (using sine, cosine). The idea of finding a solution came from a Japanese puzzle. This was my third program (by the way) made on GW BASIC (lost on floppy disks). There is a very interesting algorithm. In the cycle, rectangular ships move from left to right. If a square has always been drawn at the same point during the cycle, then it should be drawn exactly there. If a square has not always been drawn at the same point during the cycle, then it should not be there. All other meanings are not yet solved puzzle. This is where the idea of rules and layers came from. I did not study the neural network, but only now began to understand in great detail. I want to find this perfect formula. I want to do voice recognition offline. My project recognized only a few letters. I studied everything about sound. A musician himself. It turned out that it was not easy. In general, I want to make and write to a microcontroller such as ESP32. I like electronics with programming. I have some experience. This is a very interesting direction.
Hi, sure, it should be possible. I think that if you increase the frame by half (frame += 0.5) it should do the trick, but make sure to use Math.floor when you are drawing the image showVideoFrame(video,Math.floor(frame)). Then it should draw the same frame twice. Let me know if it doesn't work for some reason and I'll try to code it and send to you.
@@Radu I cant believe how fast you replied. Thanks a lot for your comment. I still have not reached the phase in my project to do this. When I do it I will let you know if is it working or not.
Hi, I don't know the answer to this... I do know that you can export a downloadable video file from the canvas, though. You can implement it from scratch or use a library like this one, for example: github.com/antimatter15/whammy But I'm not sure if audio can be easily merged into the resulting video... I quickly found this post claiming to merge audio and video, but haven't tested it myself. Hope these help.
Yeah, have to say, glad I finished with that video footage myself :-D Upcoming tutorials focus on the visual / sound effects but I don't need the video to teach those.
