235 - Pre-training U-net using autoencoders - Part 1 - Autoencoders and visualizing features

Code generated in the video can be downloaded from here:

The video summarizes the concept of autoencoders and walks you though the code for using autoencoder to reconstruct a single image. It also walks you through the code for displaying feature responses of various layer in a deep learning model.

[aggreym.muhebwa]

This is exciting. I am having trouble with improving the performance of my segmentation models. Looking forwards to Part 2. Thanks for the great work that you are doing.

[mdyounusahamed]

I wanted to take a moment to reach out and thank you for creating such informative and helpful videos.

[lion]

Just watched this and without any hesitation clicked the subscribed and like button. YOU are absolutely great. Keep it up.

[yuanchen]

It is a pretty good and clear video for explaining how to the training network and display intermediate parameters and can learn skills for tuning it in Unet pattern.

[akashravi]

You are the on of the best classic content creator for computer vision. Kindly do it for nlp as well.

[kyawnaingwin]

Thanks Sir. I cann't wait til you have uploaded the code to the github. I am typing codes from your screen. After a few typo fixes, i can see what you explained, on my screen.

[KarthikArumugham]

Great concept and this blew my mind. Never knew you could use pre-trained weights other than the ones trained on datasets like imagenet.

[umairsabir]

please start some series on mask rcnn. Thanks for your contribution to this computer vision world.

[maxmaximus]

Since the original images might be divided into 256x256 patches, when training the Encoder do you recommend also including the patches that doesn't include the region-of-interest. What about when training the Decoder. Also, whats the effect of "Smooth blending" of patches on the Encoder vs Decoder training.

[aravindangovindharajou]

Is there possible to add blocks in UNet's encoder and decoder to add new image restoration algorithm like ( grey world, retinex )

[farahhaddad]

thank you for this amazing content, do u have any idea about bushfire
satellite datasets? I can only find satellite imagery of the fires
in csv.
in csv files i can't see semantic segmentation as a result !!!!, and are fires possible for semantic segmentation ?

[shubhamshevale]

line 77 : model_for_visualization = Model(inputs = my_model.input, outputs = outputs)
i am getting error as
name 'Model' is not defined
please help me in this

[wahidullah]

thanks dear sir, could you please upload next video.

[neerajsaxena]

How much RAM your device have ? Mine crashed while trying to running this autoencoder.

[fraoney]

hi sir, i want to ask, could we use CNN with HOG and linear SVM classifier for object detection?

[shriniketankulkarni]

How can we apply u net segmentation on images which have an odd shape such as 450x1450 how do we use pachify in this case ?

[padma]

Nice video sreeni, send me link for the dataset.

[SindyanCartoonJordanian]

Would you please make atutorial about mask RCNN

[kamaleshkarthi]

@DigitalSreeni  one more question if you're free answer me without fail....
Case 1: I trained a yolo v4 model with two classes. Now i has to train same model with adding another two classes. Train the model without losses of previous two classes weight...is this possible.
My answer : reserving extra node in output layer. Can i do this?
Your answer for case1 :

Case 2 :
Dataset description: 4k images with two classes and balanced classes. Using this data set i trained two model using tiny-yolov4.
My question is:
Model 1 : trained all 4k images. 20k max_batches . getting 84% accuracy avg loss 0.12xxx
Model 2:
Cycle 1 :i trained 3k images with 20k max_batch getting 94% accuracy.
Cycle 2 : i trained 1k images with 20k max batch using last weight of cycle 1. After completion i am getting 94% accuracy. And avg loss 0.0xx.

My question is both the model i trained same set of images why result is different. Training small set of image is good?
Even though i increased model 20k+20k max batch thare is no improvement.
Note: cfg file are same for both model.
Thanks

[geethaneya]

Sir, can we use this concept for image classification problem?