Thanks for the video! Straight to the point. Just to add-up, the histograms that we extract per cell in the block should be concatenated and form the final feature vector that describe the block. Then we shift the block (window) by a stride and repeat. For example. 64px x 64px block with 8px x 8px cell and a stride of 8px. Typically, we feed the HOG feature vector to a classifier (SVM, MLP, etc.) and train it on object/non-object HOG features (The more negative examples we feed, the more robust the classifier will be and therefore we can reduce false detections/false positives). Because we are going to see multiple detections (because of the overlapping scrolling window) of the same object in the image, we use non-maxima suppression to keep only the detection with the highest probability. Lastly, to detect various sizes of the same object in an image, and since the scrolling window/block has a fixed size, we use pyramid algorithm, that essentially resize the image every time (and the object of course) before applying the scrolling window again.
why he subtract 100 -50 and not 50-100 but in the horizontal he goes for 120-70 so one time he (initial point - terminal point) and other horizontal is ( terminal point - initial point)
Those are not actually dots. Each "dot" is an arrow representing the gradient vector. That vector has a direction and magnitude. You actually won't get that image, is for display purposes only so we can get the idea.
Thanks for trying but this is too confusing. Not because of the difficulty of the material, but because of missing pieces in the explanation. You isolate a group of 8x8 pixels, compute gradients and angles and make a histogram of them for that 8x8 cell. But then what? Do you do likewise for the next group of 8x8 pixels? Unclear. Had to downvote, sorry.
very effective 12:47 minutes... thank you so much
Such a good video! Thank you. I can feel your passion through my latop. Please keep up the good work!
Man your passion just made me excited loving this series
Thanks for the video! Straight to the point.
Just to add-up, the histograms that we extract per cell in the block should be concatenated and form the final feature vector that describe the block. Then we shift the block (window) by a stride and repeat. For example. 64px x 64px block with 8px x 8px cell and a stride of 8px.
Typically, we feed the HOG feature vector to a classifier (SVM, MLP, etc.) and train it on object/non-object HOG features (The more negative examples we feed, the more robust the classifier will be and therefore we can reduce false detections/false positives).
Because we are going to see multiple detections (because of the overlapping scrolling window) of the same object in the image, we use non-maxima suppression to keep only the detection with the highest probability.
Lastly, to detect various sizes of the same object in an image, and since the scrolling window/block has a fixed size, we use pyramid algorithm, that essentially resize the image every time (and the object of course) before applying the scrolling window again.
emitting so much positive energy : )
Thanks for this video that are to the point and straight to the point
Nice Explanation,Sir.Needs More
Like this instructor!
thank u so much
thank you
U look like Leonard from big bang theory
By the way nice explanation
You look cool and your content is impressive
Thank you sir!!
just found this and the best explanation out there.!!
This was a great video your very inspirational, now I know how to apply the stuff I'm learning in linear algebra.
very good explanation sir
thanks a lot for the simplified yet effective explanation.
Glad it was helpful! thanks Gagan
nice explanation !
Thank you so much for clearing my concepts
Very informative and nice explanation sir. Thank you very much
Thank you for making this video! Very clear and helpful overview of this operator :)
Thank you so much for this video, exquisitely explained bravo!
Excellent explanation and great energyyy.
great video! very good explanation, thank you!
amazing content
thats the beautiiiiii of it
Character In the video It's great, I like it a lot $$
Very good
Hey Leonard!!
From TBBT
Sir, can you suggest a resource where the python implementation is given? Thanks!
Thanks for the nice explanation Dr. Ahmed. I was wondering if you can suggest any scientific paper or book maybe for further read.
why he subtract 100 -50 and not 50-100 but in the horizontal he goes for 120-70 so one time he (initial point - terminal point) and other horizontal is ( terminal point - initial point)
Sorry! I can't understand how did you get a dotted image 10:02 from a gradiant image. please can someone explain to me? :(
Those are not actually dots. Each "dot" is an arrow representing the gradient vector. That vector has a direction and magnitude. You actually won't get that image, is for display purposes only so we can get the idea.
Hi Prof thank you for the video, excellent explanation, how can I contact you ?
good lecture
can you help me in histogram of depth oriented gradient HOD
Where is the next exercise? :/
Thanks for trying but this is too confusing. Not because of the difficulty of the material, but because of missing pieces in the explanation. You isolate a group of 8x8 pixels, compute gradients and angles and make a histogram of them for that 8x8 cell. But then what? Do you do likewise for the next group of 8x8 pixels? Unclear.
Had to downvote, sorry.
Character In the video It's great, I like it a lot $$