TRAIN A CUSTOM YOLOv4-tiny DETECTOR USING GOOGLE COLAB ( FREE GPU )

Hello sir, thank you so much for the tutorial! I wanna ask about how many data images should we take?

Thank you for this video, However I there is an error in command15)a) when i want to train which says(AttributeError: module 'tensorflow.python.ops.control_flow_ops' has no attribute 'case') I'm using tf version 2.14.0 and i tried to use 2.13.0 but the problem was not solved

God bless you ❤

Can i use yolov4 Tiny model for web ?

for doing custom object detection using yolov8 tiny, other than cfg file changes is there anything else to be changed in the procedure?

after 1000 epochs of training, i get an error - cudnn status bad param, any solution?

Hello, i really like your work. i have a question. whenever i executes the train custom detector part, it only iterates till 1000. After the mAP calculation the process terminates with an error (cuDNN status error and cuDNN Error). Hoe can i resolve that??

Thank you so much for your tutorial! It was so helpful! I am training my custom data set as I type this. I do have a question: when I ran step [20] all of my .jpg and .txt files were unzipped into the data folder instead of the way it was in the video where the obj folder was copied and then inside of it were all of your .jpg and .txt files. Do you know why that happened? I fixed this by making a folder "obj" and moved every .jpg and .txt file into the folder. I had a very small data set of 55 .jpg and 55 .txt so only 110 things to move, but if this happens when I use a larger data set like yours then I will have to move them all one-by-one like I did for the 110. Once again thank you so much! Subbed.

Hi, thanks for this video. I have trained a V4-tiny model with the collab notebook before and it worked. But when i added more pictures into the set and rerun the training( step 10) the training stopped after 'Create 6 permanent cpu-threads '. I have changed the subdivision to 32 and tried again but still not training. Is there anything that i am missing? * I have labelled the additional images on another computer. would that make a difference to what the computer is reading? Thanks

Thank you sir, it really worked for me

You are amazing i've watched 5 other videos and yours is the only one that worked and I could understand thank you so much you've earned a sub and a like

Hello,
first thank you for this, I've trained a few models with your help.
However now i have much larger data sets which i want to train, but the training cell stops running shortly after the mAP calculation.
I have followed your tutorial correctly though.

what abt yolov7-tiny, hope u release tutor soon

Hello ,
Is there a way to check map and losses on tensorflow

possible to do that in jupyter notebook

how i can put a beep alert after detection.?possible to do a tutorial about it???

bravo! amazing honestly, two days work trying to get this working locally and now knowledgable and can see where I was getting caught up at.. got a model with 85% accuracy which is good going.
Great examples of realtime image and video at the end as well - thanks!

Hello can i run yolov4 for 300 epochs

Hello, first of all I would like to say that I really like your work. I saw that the real time part was written in JavaScript. Is it possible to run this code with external camera other than webcam? How can I run it? I would be glad if you help.

I have 1100 images which i converted in .txt file, it is executing images one by one and in my trainning it is showing 395 hours left ? how much time will it take to train ?

hey do u know how to test all image in one process? thx

It is very good. 🌹🌹🌹🌹🌹

thank you very much. it's awesome

Hlo sir ... thanks for this superb tutorial ... can I know how to convert this model to tflite model ???

Hi, Thanks for your tutorial, i have a question. I am training yolo tiny with 33 classes, my max batches is 66000 since I took 33*2000. But now has trained to 50k. avg loss is 0.8 but my mAP is 13%. Should I continue to train or do something? If I reach the end of max batches, should I increase it or stop and 1 more point in line region 30 avgIOU is always 0 .is that normal?Thankyou

You are awesome man thankyou so much for making this

when finished the training it displayed like ths as an output and also it didnt create a fyolov4-tiny-custom_last.weights file in the training folder in the drive .
so how can i solve this problem ??

Error:
Loading weights from /mydrive/yolov4-tiny/training/yolov4-tiny-custom_last.weights...Couldn't open file: /mydrive/yolov4-tiny/training/yolov4-tiny-custom_last.weights

How do I understand how to go about these steps as a novice. I have understood this implementation. But how will I be able to understand each command and run all of this on my own? Any procedure?

what is iteration and why our iteration is 6000 can someone pls explain

Hello I have a problem, the custom detector isnt getting trained but here's no error showing up either.

I tried video streaming webcam via colab, but it only open the main camera instead of front facing. Any idea how to fix it?

hi how i can know information such as (x, y , w, h) bout the detected object

hi thanks for the tutorial can we download your weights file i cannot find it on your github link

[yolo] params: iou loss: ciou (4), iou_norm: 0.07, obj_norm: 1.00, cls_norm: 1.00, delta_norm: 1.00, scale_x_y: 1.05
nms_kind: greedynms (1), beta = 0.600000
Total BFLOPS 6.790
avg_outputs = 299930
Allocate additional workspace_size = 39.25 MB
Loading weights from yolov4-tiny.conv.29...
seen 64, trained: 0 K-images (0 Kilo-batches_64)
Done! Loaded 29 layers from weights-file
Learning Rate: 0.00261, Momentum: 0.9, Decay: 0.0005
Detection layer: 30 - type = 28
Detection layer: 37 - type = 28
Create 6 permanent cpu-threads

after this training stops.
iam training on 64 images.

What should be done if want to also detect the other classes for which the model has already been trained in addition to mask and w/o mask?

Is it possible to test the trained model on a folder of images?

Thanks brother for your wonderful tutorial

Hi, are you available for building a custom object detector for my company?
Please share your email address.

Is it possible to export the trained model to a file to reuse later? If so, how?

I followed exactly your workbook however the training does not start. The code cell stops after the CPU Cores get initiated. I already checked the train.txt and test.txt files which are having the correct content. Also I already increased the max batches in the config file. Any idea what else could cause this issue?

good morning thank you so much for this tuto
i would like to know how i can deploy this project on my raspberry pi4?
is it possible?
please how ?

hello, thanks for your video. I have question. when im training my custom dataset, rewritten_bbox = 0,000000%. is this normal ?. and this prosses taking long time and always rewritten_bbox = 0,000000% . Thank you :)

Please why average loss not descrease ? my own dataset has max_batch=68000, i have trained 30k batch but accuracy mAP and best still 16.75 %. Can you help me please ? i quickly. Thank you.

Thank you very much, please how to save chart png to drive ? I train with a large data set, i need save chart continuously to analyze data. Thank you.

[yolo] params: iou loss: ciou (4), iou_norm: 0.07, obj_norm: 1.00, cls_norm: 1.00, delta_norm: 1.00, scale_x_y: 1.05
nms_kind: greedynms (1), beta = 0.600000
Total BFLOPS 6.789
avg_outputs = 299797
Allocate additional workspace_size = 12.46 MB
Loading weights from yolov4-tiny.conv.29...
seen 64, trained: 0 K-images (0 Kilo-batches_64)
Done! Loaded 29 layers from weights-file
Learning Rate: 0.00261, Momentum: 0.9, Decay: 0.0005
Detection layer: 30 - type = 28
Detection layer: 37 - type = 28
Create 6 permanent cpu-threads

I want to train on just 100 images but the training is not starting I am getting the above message

Well thank you very much for your tutorial it is awesome
I have a question please i have trained 1000 images for 2 classes using yolov-tiny i got an accuracy over then 90% but the predition is incorrect . (Always predict only one class ) even that I followed all the steps that you did and did the labeling perfectly .
I can't understand why?
Do i have to mention the class in every single image like (1-yes.jpg)
Can you please help me ?