Видео

Yes Look at this playlist. You can choose many model types. Include Detectron2 ruclips.net/p/PLdkryDe59y4bXa-1wOEAF4KljIMamhWd0&si=LGrnSMKlJ3IFYV8f

Ok. And did you succeed ? You can send me your dataset to my email and I would look into it.

No, I’m still trying to figure out if anyone has done this so I don’t need to create any software from scratch

@@russm195 I can try code the model if you send me the dataset and samples of the required result.

Hi. Please define waste images.

I am making a new video about Detr , as for accuracy it’s seems the same. As for time of training , it seems need more time of training.

Never met this error message.

Thanks. You are welcome to subscribe :)

Hi Send me your full code to my email , I will look into it

It is possible. Replace the image with collecting frames from a camera.

Yes, You can adjust the code : # If binary classification, replace `softmax` with `sigmoid`: model = tf.keras.Sequential([ base_model, layers.GlobalAveragePooling2D(), layers.Dense(1024, activation='relu'), layers.Dropout(0.5), layers.Dense(1, activation='sigmoid') # use 1 output node for binary classification ]) # Update the loss to binary_crossentropy for binary classification: model.compile(optimizer=adam_opt, loss='binary_crossentropy', metrics=['accuracy']) This update will output a single probability score for each image (interpreted as the probability of belonging to one of the two classes) Good luck

@@eranfeit tysm its work

Modify the code: # Define how many layers you want to freeze num_layers_to_freeze = 100 # Adjust this number as needed # Freeze the specified number of layers for layer in model.layers[:num_layers_to_freeze]: layer.trainable = False -> num_layers_to_freeze controls how many layers you freeze, starting from the first layer up to the specified count. Adjust num_layers_to_freeze based on your requirements. For example, setting it to 100 will freeze the first 100 layers, while the rest will remain trainable.

Thanks

Thanks. I will remember that in my next videos.

הי בדרך כלל אני כותב ומאמן מודלים בעצמי. בסרטונים האחרונים בדקתי ושחקתי עם כמה ספריות מוכנות. תודה

Yes. You should replace the source to your usb camera

@@eranfeit Thank you, Sir. Can it be used on trained models and other videos?

​@@EkaSantiFadhilah Your question is to general. You can find more tutorials for Object detection training here : ruclips.net/video/e-tfaEK9sFs/видео.html

Hi, Never tried with SSD . I believe it is a CNN architecture , so you have to "play" with the layers. Do some google search . I can see that someone asked it before : stackoverflow.com/questions/43493806/how-do-i-retrain-ssd-object-detection-model-for-our-own-dataset

Hi, Pytorch supports GPU cards with Cuda. Moreover , It also support CPU card . Try install Pytorch with this command instead : # CPU Only conda install pytorch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 cpuonly -c pytorch Eran

Thanks, You are welcome to subscribe ..

I did not understand your question . Which link does not work. The "midas\model.pt" is the name of the model after the move command which move and rename the file .

Thanks

Thanks

Thanks

No. Just run it in two sessions , while the output of session one will be the input of the second

Detection is a little bit “oldy” Try this one : ruclips.net/video/JbEy4Eefy0Y/видео.htmlsi=5LRcNbhWx2sSevd4

You are right. It is a problem in the Detectron2 . I did not manage to run in using the GPU Pytorch version. You can always install WSL2 in the Windows environment , to run a Linux on your machine

Yes, Watch the installation part (from 00:48) . You can see it was installed using Anaconda environment

By default, the images will be saved at resolution of 1024x1024, the original output size of StyleGAN. If you wish to save outputs resized to resolutions of 256x256, you can do so by adding the flag --resize_outputs.

Cuda is better for performance. You can always install the PyTorch version that supports CPU ( and not Gpu )

@@eranfeit oh thank you for eloborating. I was planning to use your tutorial for my very low league basketball club, to record our games and then automatically create highlights for it. I saw that the two 5-minute highlights already created frame images of 8,5GB. So ill probably run into Size and time issues when i record a whole game. But thanks for the tutorial, very insightful

@@GSTamer Thanks and good luck

Thanks. You are welcome to subscribe :)

Did you download the code ?

No

where to download??

Github code??

@@sayalisalkade9248 Code for this video: ko-fi.com/s/3f1fdb4316

Hi Here is the link : www.kaggle.com/datasets/slavkoprytula/aquarium-data-cots

Can you elaborate your question ?

Thank you :)

You can find a link for the instructions for this tutorial in the video description.

Thanks. You are welcome to subscribe.

There is a link for the code in the video description

@@eranfeit But there is no code to calculate the test accuration. If you know, can you give it?

I need it for undergraduate thesis purpose

Can't we just copy it from the video?

Thanks. You are welcome to subscribe my newsletter in the bottom of this link :) eranfeit.net

I did not understand your question

Thanks :)

Hi, I updated the video description with the link to the dataset

@@eranfeit thanks

@@uncover_ai You care welcome to subscribe , and share it among your friends

I believe the answer is yes. Did you try it on a football video ?

Actually no ​@@eranfeit

This is a pre trained model from Tensor hub. I believe the model was trained based on a permanent audio length , so the inference looks for audio in similar length Try to read the documentation if it returns the time value for detection. You can try to build your own audio detection.

Hi. You need learn : How to capture frames from a camera ? Build a model for detection objects that indicates theft. You need a dataset for that. If you have the dataset I can help with that. Yolo can do the job Then you need to code a prediction inference for your model That would detect the on fresh images or using a camera. I can help you with that. And at last you should code sms messaging. Eran

I think it less interesting. Do you have a dataset ?

@@eranfeit There is a dataset in roboflow, but it is not a very nice dataset. I tried it, but even if there was no accident, the algorithm perceived it as an accident. It is a bit of a difficult problem. I guessed by calculating the central point distance and angles of the cars, there were some problems here without a dataset, since the cars will be side by side in traffic, the model will guess wrong at the red light.

Thanks, I am more focused on Use cases and models. Maybe , I will take a short break for demonstrate cool Python computer vision pre-trained libraries , found it Git-hub

If you mange to follow , great.

Yes

Thank you

Thanks You are to subscribe to my monthly newsletter at this page : eranfeit.net ( go to the bottom of the first page ) Eran