Видео

Thank you, very clear and interesting explanation.

This was awesome. Thank you from a Computer Vision student in the UK!

1:54 -- what last section?

Fantastic, thank you!!!

Holy shit why doesnt this have more views

can anyone give me the correct step for the laplacian pyramid?? 1. Take the image fi from stage i. 2. filter the image fi with a low pass filter and thus create the image Li 3. Downsample the image li and thus create the image fi+1 4. Calculate the difference image hi=fi-li 5. cache hi 6. Consolidate all images hi 7. Repeat the above step n times.

such a wonderful explanation! thank you!

Can I get a lecture on SIFT?

I love how easy this is to understand. I guess there are multiple ways to accomplish a CE. In my text, its explained using XOR with NOR circuits to accomplish the same thing. Or, am I misunderstanding?

P=NP : ^ )

Much much better than the channels with millions of subscribers. I love it...❤❤

Why didnt I find this channel earlier...

Dear Professor Hany Farid, Thank you for sharing this fantastic course on image filtering! I find your explanations to be extremely clear and engaging. I have a quick question regarding linear time-invariant functions at around the 3:29 and 3:57 minute marks. The formulas shown use h[-3-k] and h[-4-k]. I was wondering if this might be a typo, as I would have expected h[-1-k] and h[0-k] to be used instead, or if I am missing something.

A viewer noticed that there is a bug at the 03:43 mark (nice catch) but I accidentally deleted their comment (sorry viewer). The code at this mark should read: M1z@M1z.T (not M1z@M2z.T)

Obvious typo on line 9 at 3:43: should be M1z@M1z.T + ...

Thanks Professor for your knowledge

Very clear explanation, thank you

This is the most comprehensive CV course I have ever seen. I was always looking for something like this but I always found general datascience courses. Thank you so much.

you is a legend

amazing point of view for explaining number of the features we want for a model won't change the cost function.

Wonderful lesson! Thank you

Amazing! Thank you

thank you for these lecture series, they are interesting

Can't be simpler. Best explanation

Thank you

Wow the legend that you are. Thank you!!!

Wow such a clear and concise video! And you make sure to review material that student should know! Amazing

Awesome explanation!! As on all the videos. Thanks for the uploads! 🙌🙌

Thank professor its really helpful

Thank you for this real world problem exercise ! I guess any rigid motion may be explained as a "combination" that ?

Why didn't anyone comment on a really good explanation?

how do you generalize this model to a dataset that "don't pass through the origin" ? Do we need to "center" the data before using it ?

Loved it!!

Great work sir. I never see any one in you tube covers in this much depth❤😊your students will be lucky to have you

Thank you professor for your effort this was really well explained, looking forward to learn more from this playlist

at 6:21, may be the summation of impulses representation is not right if the unit impulse signal is considered to be centered.

You are great at teaching. ❤

VV Good

awesome video thanks!

how comes this channel has only 500 subs very valuable stuff here

Hi Prof. Farid, At 3:50, since the "*" notation stands for convolution, shouldn't the filter be [1, -1]? Assuming right is the positive x direction The following is some python output: a = [0, 1, 0, 0] b = scipy.signal.convolve(a, [1, -1], mode='same') print(a) print(b) [0, 1, 0, 0] [ 0 1 -1 0] Similarly in 13:27, I think the d filter should be reversed.

Your videos are great! Thank you for sharing!

What is the difference between grayscale intensity vs luminance from YCbCr?

This was a great video thank you

Thank you!

Thanks from Türkiye.

Hi Prof. Farid: I was having great trouble of understanding what sepfir2d do, since intuitively we are talking about convolving but here we are using a different function(sepfir2d). Since skimage API doesn't provide much details, I have to try and guess for a few hours to really understand the the linkage between the two concept. I felt it would be better to show the code using signal.convolve2d, then show the equivalent but more computational efficient form would really help future student. The following python code is not a bulletproof test but enough to show my understanding is correct : import numpy as np from scipy import signal import random PRE_FILTER = np.array( [0.030320, 0.249724, 0.439911, 0.249724, 0.030320], dtype=np.float32, ) FIRST_ORDER_DERIVATIVE_FILTER = np.array( [-0.104550, -0.292315, 0.000000, 0.292315, 0.104550], dtype=np.float32, ) KERNEL_FIRST_ORDER_DERIVATIVE_FILTER = np.zeros((5, 5), dtype=np.float32) KERNEL_FIRST_ORDER_DERIVATIVE_FILTER[2, :] = FIRST_ORDER_DERIVATIVE_FILTER KERNEL_PRE_FILTER = np.zeros((5, 5), dtype=np.float32) KERNEL_PRE_FILTER[:, 2] = PRE_FILTER def test_sepfir2d_convolve2d(test_count: int): for _ in range(test_count): random_image = np.array(random.sample(range(0, 255), 100)).reshape(10, 10) sep_gradient_x = signal.sepfir2d(random_image, FIRST_ORDER_DERIVATIVE_FILTER, PRE_FILTER) conv_gradient_x_row = signal.convolve2d(random_image, KERNEL_FIRST_ORDER_DERIVATIVE_FILTER, mode="same", boundary="symm") conv_gradient_x = signal.convolve2d(conv_gradient_x_row, KERNEL_PRE_FILTER, mode="same", boundary="symm") print(f'sepfir2d and convole2d are equal: {np.array_equal(sep_gradient_x, conv_gradient_x)}') test_sepfir2d_convolve2d(50) returns all true for 50 random tested image

These videos are so good and clear, how come no one leave a comment?

your videos are fantastic... I'm happy I found your channel ❤