I’d like to do some image processing, image detection, and motion detection using machine learning for lots and lots of sporting broadcasts.
Is there anything wrong with this Acer gaming computer for that purpose?
4050 only has 6GB worth of GDDR GPU memory.
It’s perfectly serviceable for image processing even for most CNN-based models, as long as you are not looking to train the models.
If you want to train with a large dataset though, I think 11GB is the bare minimum.
2 Likes
Is GDDR GPU memory the graphics card RAM size?
I want to collect data from game tape. Does that require training large datasets?
Yes. It also isn’t the fastest interface. However, HBM would be way more power hungry on a laptop.
If there are pretrained models already capable of doing what you want to achieve, then you don’t need to train your models.
Also, if you need to extract time series causal relations from a gameplay video, you might need more than just a image classification or object detection model.
What I’m trying to do is this:
He does it with a CPU, but the game film he loads are 20MB, and the ones I have are 7GB. When I tried loading it using Python, it took to friggin’ long.
I could ask my sports betting and analysis clients what they use for post match data they sell to clients. For live in play its not AI but humans.
Some of my clients are doing over 500,000 events a year. They are also using blade servers that hold multiple GPU’s.
1 Like
cool
For pure object detection inference, I think 4050 would be fine.
There’s really nothing to track between frames, so the model is basically looked at a frame at a time.
Looks like there is some training going on, although the training dataset is rather small in comparison to the validation. I haven’t watched the entire video.
I think it’s better for me to put off buying until I get a better handle on what’s going on. I really want to buy by the end of the year because it’s 5% cash back on Amazon this quarter, but it’s imprudent at this juncture.
The Yolo on my CPU takes an hour to detect an label one minute of tape, goodness gracious.
Do you have a specific computer you recommend?
yolov5.pt is only 175MB. I don’t know what your training dataset is, but I doubt it would be even close to the level of imagenet. So 6GB GDDR on that 4050 should be sufficient.
Also, I still really doubt you would need to retrain your yolo just to detect people and a ball. Yolo does that really well out of the box.
Alright I just ordered it.
As I watch more of the video, I encounter more issues. He doesn’t have a GPU so he trains on Google collab. You get two hours a day on there.
Google collab has 12 GB DDR on its free tier.
For this type of inference likely any recent Nvidia graphics card will do, 6GB or larger. If in doubt get one with more memory though, like 8GB or more on a PC.
1 Like
Thanks. I think the one I’m looking at has 8GB RAM and 6GB VRAM. Am I mistaken?
I’m pretty ignorant on this stuff. I got the VRAM mixed up with the system RAM.