Hello again, I have a random question. So I am trying to train on Yolov4 following all your tutorials and its working as intended! However I just cant train with a width and height of 1024. My settings are Batch = 64 Sub = 64 Width = 1024 Height = 1024 Running on a 3060 12G. 16G of Ram and 16 core AMD 3650x I am told I need to have 1024 width and height as that is what the pre-trained / Images are on. What I get is 1472 x 1472 Create 6 permanent CPU-threads Error: cuDNN isn't found FWD algo for convolution I have looked everywhere but I don't see anyone training at 1024 even though that is what my images are on. What I also notice is that my GPU Memory goes to 11.4G - 11.6G of the 12G and then the Error appears. Would love some help. Thanks UPDATE - Ok so I just noticed that my compute_capability is only 860.
@@TheCodingBug With your permission, I would like to ask one more question. I am using jetson nano and fps is very important to me. I want to use tensorflow lite. Which YOLO version do you think would be more useful for me?
Hi Mr TheCodingBug. I like a lot your channels for all the well explained content that you were creating all this time. Now, back to this video... what is the hardware specs that you used to test the 7 models?
Thats interesting. I have not compared these two myself so I cannot say for sure. As per research paper, the YOLOv7 is faster and have higher mAP on COCO dataset. The darkent based base YOLOv4 was giving an FPS of up to 13 (ruclips.net/video/FE2GBeKuqpc/видео.html), while base YOLOv7 is giving up to 16FPS. So it's some improvement there that we can notice, keeping in mind that darknet is based on C++ and requires a lot of effort to setup.
Can u try demonstrating a custom dataset training in kaggle? Kaggle is better than google colab as it gives u 30 gpu hours per week and doesn’t restart like google colab Also can u show us how to implent Yolov7-tiny and a python code demo?
Python demo for all variations of YOLO v7 can found here: ○ YOLOv7 Custom Object Detection (Win, Linux): ruclips.net/video/-QWxJ0j9EY8/видео.html ○ YOLOv7 Complete Tutorial (Colab): ruclips.net/video/_CkXDjmT8dc/видео.html ○ YOLOv7 Complete Tutorial (Windows, Linux): ruclips.net/video/n2mupnfIuFY/видео.html I will try custom object detection on Kaggle.
God job! Make a compare with others yolos !!!!!!
Hello! I'm really curious about the resolution information for these videos. Thank you for your help!
All videos are 480P.
THX！@@TheCodingBug
Thanks for sharing~!
On which device did you get these fps values?
Hello again, I have a random question. So I am trying to train on Yolov4 following all your tutorials and its working as intended! However I just cant train with a width and height of 1024.
My settings are
Batch = 64
Sub = 64
Width = 1024
Height = 1024
Running on a 3060 12G. 16G of Ram and 16 core AMD 3650x
I am told I need to have 1024 width and height as that is what the pre-trained / Images are on.
What I get is
1472 x 1472
Create 6 permanent CPU-threads
Error: cuDNN isn't found FWD algo for convolution
I have looked everywhere but I don't see anyone training at 1024 even though that is what my images are on.
What I also notice is that my GPU Memory goes to 11.4G - 11.6G of the 12G and then the Error appears.
Would love some help.
Thanks
UPDATE - Ok so I just noticed that my compute_capability is only 860.
try batch 32
Are you running your test using an IP Camera or an mp4? yolov7 on my machine is fast only with my webcam
Mp4 file.
please how can i improve my fps my yolov7 model when i run it with my custom dataset show 4fps and it is really low what can i do ?
In your codes, did you use any framework especially like tensorflow lite?
In this video, no. But I have yolov4 video on my channel where I convert yolov4 to tflite and use it in mobile application.
@@TheCodingBug thank you for your feedback
@@TheCodingBug With your permission, I would like to ask one more question. I am using jetson nano and fps is very important to me. I want to use tensorflow lite. Which YOLO version do you think would be more useful for me?
@@senayuksel3342 yolov8 nano or yolo-nas
Hi Mr TheCodingBug. I like a lot your channels for all the well explained content that you were creating all this time. Now, back to this video... what is the hardware specs that you used to test the 7 models?
I am using GTX1060 6GB for all of these runs.
cool video, though i am not sure which is better here?
It depends on the input. From top view, larger models are better.
From street view, YOLOv7 is better due to FPS.
can u show how to display fps on screen? thx
I was speaking with some people and they claimed that Yolov7 is not as good as Yolov4. I think I will stick with Yolov4 myself.
Thats interesting. I have not compared these two myself so I cannot say for sure. As per research paper, the YOLOv7 is faster and have higher mAP on COCO dataset.
The darkent based base YOLOv4 was giving an FPS of up to 13 (ruclips.net/video/FE2GBeKuqpc/видео.html), while base YOLOv7 is giving up to 16FPS. So it's some improvement there that we can notice, keeping in mind that darknet is based on C++ and requires a lot of effort to setup.
@@TheCodingBug Yeah I think that is the issue, the setup for YOLOv4 and Darknet. Will you be doing an extended tutorial on training with Yolov7? :)
@@MEATHEADBooYA Yes. I am working on custom dataset tutorial for YOLOv7.
we need custom object detection on python 3.9 and last version tensorflow please you haven't like video !
How can you demonstrate the FPS values ?
Higher FPS means the model variation is faster.
I mean how you were able to show these fps values on window ?
?
@@pearcexx6085 Modify detect.py as mentioned in my other tutorial: ruclips.net/video/_CkXDjmT8dc/видео.html
Appreciated 🙏🏻
Can u try demonstrating a custom dataset training in kaggle? Kaggle is better than google colab as it gives u 30 gpu hours per week and doesn’t restart like google colab
Also can u show us how to implent Yolov7-tiny and a python code demo?
Python demo for all variations of YOLO v7 can found here:
○ YOLOv7 Custom Object Detection (Win, Linux): ruclips.net/video/-QWxJ0j9EY8/видео.html
○ YOLOv7 Complete Tutorial (Colab): ruclips.net/video/_CkXDjmT8dc/видео.html
○ YOLOv7 Complete Tutorial (Windows, Linux): ruclips.net/video/n2mupnfIuFY/видео.html
I will try custom object detection on Kaggle.