Learn Computer Vision using AI: kevinwoodrobotics.com/product/computer-vision-using-ai/ Code and Doc: kevinwoodrobotics.com/product/ai-feature-detection-and-feature-matching-with-dinov2-from-meta-ai/
@@kevinwoodrobotics That's great, thank you. I was just curious if the training models are too large to run locally, but it looks like DINOv2 is doable at around 5GB. Some of these models are enormous in size and I'm unable to test locally.
I want to extract features from an image using DINO V2. Is this method applicable? I need to calculate the cosine similarities between two images using DINO V2 extracted feature
Really helpful video. I have images that are not of 224 x 224 sizes, and are of huge dimensions (7000 x 8000 as an example), and not all are squares. Will dinov2 work then
@kevinwoodrobotics Does resizing the image affect its original dimensions or cause any loss of important information? Specifically, does dinov2 preserve the original size and maintain the aspect ratio after resizing, or will the resized image lose any critical functionality?
@@kevinwoodrobotics Thank you so much for your reply. if you could make a video on finetuning, it would be a great help. I searched on youtube a lot, but I couldnt find a good video, which can nicely describes Dinov2 fintetuning steps.
Learn Computer Vision using AI: kevinwoodrobotics.com/product/computer-vision-using-ai/
Code and Doc: kevinwoodrobotics.com/product/ai-feature-detection-and-feature-matching-with-dinov2-from-meta-ai/
Nice video, i just have one question,can we use dinov2 model for face regonistion?
yeah i think you could extend it to that application
And do we need fine tuning for this task? Or we can proceed with pretrained dino
Do you have everything installed locally for DynoV2 or are you running the final
scripts online on something like huggingface?
Everything is run locally with modules installed in my env. I’m not using the dino repo or hugging face but you could do that as well
@@kevinwoodrobotics That's great, thank you. I was just curious if the training models are too large to run locally, but it looks like DINOv2 is doable at around 5GB. Some of these models are enormous in size and I'm unable to test locally.
@@vizdotlife yeah the 5gb one is a good option to use for most hardware!
@@kevinwoodrobotics I wonder if DINOv2 could also do well in background removal using semantic segmentation.
I want to extract features from an image using DINO V2. Is this method applicable? I need to calculate the cosine similarities between two images using DINO V2 extracted feature
Yes this would work for you
@@kevinwoodrobotics Here you did feature marching, but what is need is the code for the feature extraction so that I can study it and Understand.
@@AbduljaleelAdejumo you need to do feature extraction before you can do feature matching. My code does both
@@kevinwoodrobotics@kevinwoodrobotics Thanks. Can you give a GitHub link to the code?
Really helpful video. I have images that are not of 224 x 224 sizes, and are of huge dimensions (7000 x 8000 as an example), and not all are squares. Will dinov2 work then
Yes it should work
@@bluegrey2468 I just resized it so it can run faster
@kevinwoodrobotics Does resizing the image affect its original dimensions or cause any loss of important information? Specifically, does dinov2 preserve the original size and maintain the aspect ratio after resizing, or will the resized image lose any critical functionality?
thank you for this video. can you please make a video on finetuning dinov2 on custom data, e.g. medical images. thank you so much
I can add that to my list
@@kevinwoodrobotics Thank you so much for your reply. if you could make a video on finetuning, it would be a great help. I searched on youtube a lot, but I couldnt find a good video, which can nicely describes Dinov2 fintetuning steps.