Detecting Facial Keypoints with Deep Learning | a very simple top 5 kaggle solution

[@AladdinPersson]

Hey let me know which Kaggles could be interesting for future videos and what you think of these so far. Share your thoughts on what you think can be improved with these Kaggle videos so I can make them even better :) Timestamps for the video: 0:00 - Introduction to competition 0:52 - Why? 1:22 - The data and the task 3:13 - Metric used 3:43 - A more in depth look at the data 7:24 - The steps of the solution 10:00 - Code: config 13:00 - Code: dataset 18:52 - Code: split_data 20:32 - Understanding submission format 21:57 - Code: utils 26:07 - Code: train/main file 31:22 - Fixing a small error 32:05 - Checking the score of submission 32:40 - How to improve it further? 34:55 - My 2 things to try for improving the score 35:15 - Ending & my thoughts on Kaggle

[@rohangupta9548]

Something related to self driving It will be helpful

[@maxscheijen]

I really like this series about kaggle solutions!

[@benmiss1767]

Yup I agree it’s really great !

[@yehdhihanna6114]

Another one. I mean i agree too :)

[@frkkful]

Thank you. As a ML beginner, I learned a lot from your approach.

[@HailemariamMogesT]

WoW! Great Work! I'm knew to computer vision but want to delve into in practice. Do you have any idea how can I get a facial key point detection dataset? AND how to combine the FER2013 dataset with facial key point detection after getting facial key point detection dataset to do the facial expression recognition? Thank you.

[@aurkom]

This series of kaggle competitions is quite fun!

[@ggzet6535]

Thanks for the videos! How is your color theme calls?

[@sumanmahapatra8881]

Aladdin, please continue implementing some of the GAN models. Please implement Encoder Decoder based GAN model involving a Latent space like BiGAN ( Adversarial Feature Learning). Interpreting latent space is a burning topic now a days. Thanks.

[@alexkonopatski429]

This is such a great series and i really like it. But how did you get these files like "15-Keypoints", "4-Keypoints"? Did you extract them and or how did you do it. Because if i download the dataset from kaggle and i don't get these files. I looked through your code i cannot find it where you did it. It would be pretty cool if someone else could might say it or you. I wish everyone a good day!

[@murtazaburhani4022]

I thank youtube algorithm for dropping me on this channel! It's truly a gem.

[@deepamgautam]

Can u plz make a video on landmarks (keypoint) detection in 3D images, for example 3D MRI scan images

[@for-ever-22]

These videos are extremely helpful. Thank you

[@Jake-om9no]

At 31:05, you said: "I trained it for 10 epochs, I first had the learning rate to be 1e-4, and changed it to 1e-5, then I trained for a couple of epochs more..." Given what you said above ↑, just wonder (perhaps a stupid question): how did you manage to stop the training before changing LR to 1e-5 and continue the training after that? I downloaded your code, but it does not seem to provide a mechanism to do that. (Though I can change the code to perform the task, I just wonder how exactly you did do it, )

[@macaronminmin]

Thanks for the amazing video. unrelated but I just love your keyboard sound! Could you share what keyboard you use?

[@AladdinPersson]

How did you even hear it? I bought with silent mx reds in order for it not to make sound haha

[@ardavanmodarres4720]

good job! thank you so much. you are the best, man!

[@rs9130]

how to extend for multi-face detection. is there any dataset?

[@MrMarcowally]

if you are using effnet b0 from pretrained why not use the mean and std values used in ImageNet pretraining to normalize? the values you use, you calculated from your dataset?

[@motbus3]

Hey. I worked quite q but with such applications. The images are not duplicates but different facial expressions

[@emransaleh9535]

Good job! It would be nice if you add a github repo of this work in the description.

[@AladdinPersson]

Added now: github.com/aladdinpersson/Machine-Learning-Collection/tree/master/ML/Kaggles/Facial%20Keypoint%20Detection%20Competition

[@emransaleh9535]

Perfect. Keep on the great job!

[@datakube3053]

Will you please tell how to improve coding skills and logics for Deep learning

[@truonggianga2tk42]

Thank you very much and wait other video about kaggle competitions :)

[@AdobadoFantastico]

Thanks, this is super enlightening.

[@andihaki]

Any "working" references for training face keypoint in tensorflow? Thank you

[@chinweikechukwu4781]

Hi, Aladdin. Please do a video on Grad-Cam

[@yannickpezeu3419]

Hi Aladdin, would you do a tutorial on Tensorflow Probability one day ?

[@1chimaruGin0_0]

Awesome video. If possible, please make videos of ensemble and stacking.😁

[@mohdkashif7295]

Please make a series on using pretrained models in pytorch using basicblock and bottleneck, because I have loved your series of implementing models but i want to see pre existing models modification according to data.

[@NatureSerenades654]

I trained on gtx1070, but found that the val loss just don't decrease after a few epochs, and I changed train.py , in detail, I only let the .pth file update when the val loss in this epoch is lower than the lowest val loss. However, my submission only got score 4.9. Could you help analyse this ? Thanks! Poor English, hope my description is clear.

[@odysseashlap]

This is top! Keep up the good stuff!

[@pabbirahul7495]

hello sorry for this interruption but sir please check the ESRGan tutorial comments please upload the pretrained weights for them which helps us a lot thank you

[@adamtran5747]

I cannot find images used in this competition. Did Kaggle take them off?

[@hackingpirate3973]

can stream the watermark remover using GAN on twitch sometime? BTW great video, liked it!

[@samibenhssan3121]

Please @aladdin keep going with kaggle solutions series but with more recent competitions....

[@AladdinPersson]

Sure, got any recommendations?

[@samibenhssan3121]

@@AladdinPersson What about this one www.kaggle.com/c/siim-covid19-detection

[@mythorganizer4222]

@@samibenhssan3121 bro it's ongoing 😂

[@oliverc3104]

could you please do a neural turing machine from scratch? That would be cool.

[@chedychedychedy]

how do you extract the images from csv files?

[@angelospapadopoulos7679]

import numpy as np import pandas as pd import os from PIL import Image def extract_images_from_csv(csv, column, save_folder, resize=(96, 96)): if not os.path.exists(save_folder): os.makedirs(save_folder) for idx, image in enumerate(csv[column]): image = np.array(image.split()).astype(np.uint8) image = image.reshape(resize[0], resize[1]) img = Image.fromarray(image, 'L') img.save(save_folder+f"img_{idx}.png") csv = pd.read_csv("test.csv") extract_images_from_csv(csv, "Image", "data/test/")