Building a GAN From Scratch With PyTorch | Theory + Implementation

[@stuartallen2001]

Wow awesome video! I like this style where you give the intuition for a project first and then go about implementing it. I hope there's more like this in the future!

[@AssemblyAI]

Awesome, thank you for the feedback! -- Patrick

[@aramsargsian7182]

😂

[@tiwaritejaswo]

no explanation, he's just reading the code lines aloud.

[@luluw9699]

Do you even code? If you know pytorch and pytorch lightning, you could understand the code he wrote. Besides if you want in depth knowledge you can get from other videos and follow this one.

[@tianbowen721]

It's my second time watching it, I"ll make sure to watch again for better understanding

[@mamase23]

For anyone getting errors, related to `unexpected keyword gpus` or related to manual optimization, consider downgrading your version of `pytorch-lightnning` to 1.5.10. When he shows the documentation he's following it's hard to tell but it's in the top right corner. Heaven forbid he adds that to the description or in one of the many previously asked questions...

[@mohammadarabzadeh1219]

it's "devices" now, change 'gpus' to "devices"

[@Prabudh-q5j]

Thanks very helpful!! I actually poasted a question about the same i was just checking if he replied and then you stumpled upon 🙂🙂

[@NehaSharmabcs]

can u share the whole working code ?

[@Chia_seeds]

@@mohammadarabzadeh1219 no it cannot work

[@TheCreativeautomaton]

super clear pytorch GAN example

[@djsosbxbdirndxnkcbebxhxbe]

What is the point of just reading already written code first and writing new code while reading it then?

[@julio11789]

Callback.on_epoch_end has been removed in v1.8. I replaced that line with on_train_epoch_end

[@TuanNguyen-su5ty]

awesome video, this video make me alignment more to GAN

[@junjuljunjul4392]

Excellent video, but has the full notebook been posted somewhere?

[@saqibrajaputtv3609]

sir please do create video on diffusion which will describe model through practical using pytorch or tensorflow like this

[@DeepakSingh-ji3zo]

Excellent!! Can you please update the colab link. the one in the description does not contain full code.

[@Mohamm-ed]

Thanks a lot. Could you please make another tutorial about GAN for time series.

[@user-ug6kk5ux5q]

why do you generate random data twice? you have a variable self.validation_z = torch.randn(6, self.hparams.latent_dim), that you only use in the plot function wtf. And durring training you generate other z values wtf. And you don t have a validation function. I don't understand what is happening in here

[@rohitray5947]

when I tried to train....an error occurred: TypeError: Trainer.__init__() got an unexpected keyword argument 'gpus'.....how to fix it?

[@vineethanair-ih4dl]

i got the same error. how did you fix it

[@NehalVerma-zr4mq]

I too got same error

[@user-ug6kk5ux5q]

it is because the gpus param was deprecated. Write: trainer = pl.Trainer(max_epochs=20, accelerator='gpu', devices='auto') instead.

[@giannissavvas7444]

Awesome video! Which is the version of pytorch lightning you were using at that time?

[@unknownhero6187]

Thank you! As I know that dude also have a great python channel

[@AssemblyAI]

haha, yes he has :)

[@patloeber]

🤗

[@germanogomes1097]

This tutorial is amazing. Easy steps, nice code review. Even a beginner could follow along all the steps. Still I reproduce every code twice I'm getting this error 'GAN' object has no attribute '_backward_hooks'. Could help me to solve this?

[@ericbader7998]

The GColab code only includes the classes for the generator and discriminator, it looks like the rest of the code is missing or was deleted?

[@thienphanuc9939]

Thanks for a great video. However, when I run the last code I get an error. MisconfigurationException: No `configure_optimizers()` method defined. Lightning `Trainer` expects as minimum a `training_step()`, `train_dataloader()` and `configure_optimizers()` to be defined.

[@Dee-my8bg]

Did you fix the error, and if you did, how?

[@animationtv1908]

I have the same error as well

[@NehaSharmabcs]

can anyone share the whole working code please?

[@liuwendong9223]

thanks very much

[@zainmalik457]

how to use GAN model for Multi modal image data like RGB , Depth and infrared images ?

[@mr10011994]

Hope you can see this, but I have the error right at the end: __init__() got an unexpected keyword argument 'gpus' I did everything exactly you show here and check it twice... Wondering how I can reach out to solve this

[@МарияМакарова-п3й]

parameters were changed, so you need to use this instead trainer = pl.Trainer(max_epochs=20, devices=AVAIL_GPUS, accelerator='gpu')

[@ajaykumar-rv5jl]

is your code now working ?

[@МарияМакарова-п3й]

@@ajaykumar-rv5jl I could not set up environment for it, so it does not :(

[@VASUDHARANIPATHEDAMIS]

@@МарияМакарова-п3й hey now I have another error Training with multiple optimizers is only supported with manual optimization. Set `self.automatic_optimization = False`, then access your optimizers in `training_step` with `opt1, opt2, ... = self.optimizers()`. what should I do??

[@shellman-ee9bh]

​@@VASUDHARANIPATHEDAMIStry downgrading my pytorch lighting version to 1.6.4 pip uninstall pytorch-lightning pip install pytorch-lightning==1.6.4

[@fizzleee]

does the optimizer index solution still work with newer lightning versions? I get the following error "Training with multiple optimizers is only supported with manual optimization"

[@____zero_____]

had this issue too. Just revert back to the version in the video (1.5.10) by using "pip install pytorch-lightning==1.5.10" to specify the version

[@agenticmark]

reading code isnt teaching my man. i understand whats going on here, but thats after a few years in ML. its probably not that useful for newbs trying to figure this stuff out.... i use gans for audio generation and upscaling. works great! for those interested, this is how RVC voice cloning works...

[@user-ug6kk5ux5q]

newbs should probably start with the basics like all of us did..and not jump directly to generative ai

[@gowthamiv8196]

Very Good explanation on creating GAN.Thank you, Could you please explain me how image compression can be performed using GAN?

[@vijaygill1526]

I keep getting an error "training_step() missing 1 required positional argument: 'optimizer_idx'" If I set optimizer_idx=0 or =1 then it works and I've tried upto 65 epochs, doesn't seems to change the output at all still the exact same noise image which I get before any training.

[@utkarshtamrakar3873]

i can sucessfully run the code till trainer (model, dm) ...... but in last no display of maps after each epoch

[@michil.1192]

same here

[@bloodang6793]

in GAN class change the name of the method: "def on_epoch_end(self):" to "def on_train_epoch_end(self): self.plot_imgs()"

[@sergioruppel7269]

I have a question, im trying to do a GAN to generate new data but from 32x32 data, ive seen many Docs about changing the parámeters in the generator and discriminator, but My gan fails to train. How can i modify this one to fit those picture sizes? The data i'm using si the htru1 dataset with pulsar data. Or can You point me to something to read? Thanks

[@shrill8239]

Isolate and test run your train function and add code to trouble shoot errors at various points during training

[@ifthikhanmohamed1681]

RuntimeError: Training with multiple optimizers is only supported with manual optimization. Set `self.automatic_optimization = False`, then access your optimizers in `training_step` with `opt1, opt2, ... = self.optimizers()`. how to fix this

[@en-iyi-benim]

modify your GAN class like that "class GAN(pl.LightningModule): def __init__(self, latent_dim=100, lr=0.0002): super().__init__() self.automatic_optimization = False"

[@ifthikhanmohamed1681]

@@en-iyi-benim TypeError: GAN.training_step() missing 1 required positional argument: 'optimizer_idx'

[@OmarKadry-jy5gs]

still the same error@@en-iyi-benim

[@user-ug6kk5ux5q]

@@en-iyi-benim not enough

[@Democracy_Manifest]

This is an excellent video but you really should update it. It's only 1 year old but the code is generating this error: TypeError: Trainer.__init__() got an unexpected keyword argument 'gpus'

[@pedroserrano5479]

that's probably because the more recent versions of PytorchLighnting no longer have that parameter. To stick to the tutorial you can revert the version back to around the time this was made. So you can run: ! pip install --quiet "ipython[notebook]>=8.0.0, =1.8.1, =0.7, =1.4, =2.0.0rc0" that way you can run the code from the tutorial. You might also have to replace/remove the on_epoch_end() method

[@en-iyi-benim]

modify trainer like that "trainer = pl.Trainer(max_epochs=20, accelerator="auto")"

[@zakenmaru8612]

I did devices in lieu of gpus and it worked.

[@InPoeWeTrust]

I LOVE YOU

[@enveripar8359]

How can I add custom dataset into this script

[@Dhru-va]

thank u

[@priyankvenkatesh877]

Hello, I'm trying to run the training but I end up getting this error: RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn Is there something I'm missing?

[@victoriariosvazquez2924]

Same here. Could you solve this issue?

[@CristinaJderiu-kn7gz]

And shame tv nee inteork kik nonoem come to eve tochb hatem y bldy scarat zgarie kil ti said shti it Interpol kil it and will all kisng atfi no kid toy oh epididymis tl y ring atfi eve toch. Interpol said curvok

[@VASUDHARANIPATHEDAMIS]

Hello I have an error at the end could you help Trainer.__init__() got an unexpected keyword argument 'gpus'

[@vineethanair-ih4dl]

yeah same here. did you get the solution?

What is used in the new version of Lightning is something like pl.Trainer(max_epochs=20, accelerator='auto', devices='auto')