If you want to edit videos…I do a round about things…
I take the screen recording.
Upload it to OneDrive.
Go to my laptop or PC and right click on the video and edit with Microsoft ClipChamp which is a video editor.
Then save it and either share from there or my phone…if you share from the Android phone it offers this…but you must upload your video to Google photos…which I reluctantly do… sometimes.
Sharing on a video in “Google photos” app gives these options…I use create a link…then post the link on here.
the highlighted numbers are:
mAP50 (Mean Average Precision at IoU 0.5 — core accuracy metric) = 0.765 or 76.5% accuracy so far need 80 or better is desired…really 90 or better.
mAP50-95) (Stricter, average across IoU 0.5 → 0.95 — challenging benchmark) = 0.417.
If mAP50 passes 0.80+ and mAP50-95 reaches 0.45–0.5, this model may already be viable for deployment!
I plan on having to retrain and will adjust my parameters and run it at 200 epochs instead of 100. it seems to have plateaued a bit.
Sorry I didn’t tag you, and also sorry it’s not a better video I redone it a bunch of times due to being too long or didn’t like what I did get… Hopefully with how I have settings set it will get views.
Sorry I’ve been not posting so much lately…been working on this ai…ever since I finished the image bounding boxes… I’ve had new energy to get this done.
Amazing work brother, i would not be able to do anything like what your doing. I can’t wait for the app to take off and you start getting tons of subscribers. I wish you luck, you deserve it brother for real.
Background:
I am able to train fully…however to get a good training…I need about 200 epochs…or more each takes about 39 minutes on my PC.
My PC is over locked and water-cooled, with a ton of ram…however…my video card is a Nvidia Quadro m4000…great for solidworks…bad for gaming and not for Ai stuff. So I must use the CPU and not the GPU which could be 10x to 100x faster depending on the new ai graphics card…oh yah they are $2500-25000.
So I must deal with it by ensuring I can monitor the entire training. So I can evaluate throughout the training process to stop if necessary to adjust settings and get going again ASAP. Since the final training will take 3+ days…I don’t want to wait 3+ days to find out it’s not right or way off…this way I can stop when needed to save time…but stop logically and intelligently based on evidence…
So that is where all of my errors have been…not only training but also logging everything into a web browser page that is directly linked to my training folder and has everything there for me…this is TensorBoard…it’s not uploading to a service…it’s based on my PC and has it’s own server for me to log into via the MS Edge browser.
It trains the first epoch and either doesn’t log the first epoch to TensorBoard for evaluation and moves to the 2nd epoch, or the after the first epoch it crashes in the gather charts, graphs, and test pics to send to TensorBoard and crashes.
Frustrating to say the least…39 minutes…wait and fails and start again ASAP…again and again until I get out the bugs.
Just amazes me how smart some people can be, like what you’re doing is awesome to follow, like you’ve created something people buy and help them, and you want it to be the best or can be, and that take A LOT of patience time and energy, and you’re going to deal with plants soon too…you are a mad man but kudos to you and all the success coming your way.
I have figured out I need to upgrade my videocard(GPU - graphic processing unit) to get the bugs taken care of…until the card gets here…found a nice one on eBay… I’ll continue with doing what I have been to get it running on just my CPU…and over clocked Intel i7 to 4.3ghz…still not good enough…bought a Gigabyte Nvidia RTX 3060 with OC 12gb gddr6…my per epoch times should go from 45 minutes down to 6-8 seconds per epoch to put this upgrade into perspective…plus I can just code straight forward to the GPU…vs…forcing it to work on just a cpu…and not an Ai CPU either.
I hope to have the script finished by the time the card gets here…I have built in settings to change if I had ended up upgrading…glad I was forward thinking on that.
