Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Imagen, a text-to-image diffusion model

gweb-research-imagen.appspot.com

Post reply on HN
Print view View original

331–340 of 661 posts

Re: Imagen, a text-to-image diffusion model

#331

Post by pshc »

tines wrote: post #42
daenz wrote: post #7

>While we leave an in-depth empirical analysis of social and cultural biases to future work, our small scale internal assessments reveal several limitations that guide our decision not to release our model at this time. Some of the reasoning: >Preliminary assessment also suggests Imagen encodes several social biases and stereotypes, including an overall bias towards generating images of people with lighter skin tones…

This raises some really interesting questions. We certainly don't want to perpetuate harmful stereotypes. But is it a flaw that the model encodes the world as it really is, statistically, rather than as we would like it to be? By this I mean that there are more light-skinned people in the west than dark, and there are more women nurses than men, which is reflected in the model's training data. If the model only gener…

I think it is problematic, yes, to produce a tool trained on data from the past that reinforces old stereotypes. We can’t just handwave it away as being a reflection of its training data. We would like it to do better by humanity. Fortunately the AI people are well aware of the insidious nature of these biases.

Re: Imagen, a text-to-image diffusion model

#332

Post by slg »

rpmisms wrote: post #323
slg wrote: post #318

Earlier quoted context omitted.

What makes you think those are the only options? Why can't we have an option that the model returns a range of different outputs based off a prompt? A model that returns 100% of nurses as female might be statistically more accurate than a model that returns 50% of nurses as female, but it is still not an accurate reflection of the real world. I agree that the model shouldn't return a male nurse 50% of the time. Yet a…

So, the model should have a knowledge of political correctness, and return multiple results if the first choice might reinforce a stereotype?

I never said anything about political correctness. You implied that you want a model that "provides a reflection of reality". All nurses being female is not "a reflection of reality". It is a distortion of reality because the model doesn't actually understand gender or nurses.

Re: Imagen, a text-to-image diffusion model

#333

Post by benreesman »

dougabug wrote: post #329
benreesman wrote: post #273

I apologize in advance for the elitist-sounding tone. In my defense the people I’m calling elite I have nothing to do with, I’m certainly not talking about myself. Without a fairly deep grounding in this stuff it’s hard to appreciate how far ahead Brain and DM are. Neither OpenAI nor FAIR ever has the top score on anything unless Google delays publication . And short of FAIR? D2 lacrosse. There are exceptions to such…

This characterization is not really accurate. OpenAI has had almost a 2 year lead with GPT-3 dominating the discussion of LLMs (large language models). Google didn’t release its paper on the powerful PaLM-540b model until recently. Similarly, CLiP, Glide, DALL-E, and DALL-E2 have been incredibly influential in visual-language models. Imagen, while highly impressive, definitely is a catch-up piece of work (as was PaLM…

I agree that it’s still a jump ball in a rapidly moving field, I was saying Google is far ahead, not that they’ve won.

And I don’t think whatever iteration of PaLM was cooking at the time GPT-3 started getting press would have looked to shabby.

I think Google crushed OpenAI on both GPT and DALL-E in short order because OpenAI published twice and someone had had enough.

Re: Imagen, a text-to-image diffusion model

#334

Post by benreesman »

meowface wrote: post #308
benreesman wrote: post #273

I apologize in advance for the elitist-sounding tone. In my defense the people I’m calling elite I have nothing to do with, I’m certainly not talking about myself. Without a fairly deep grounding in this stuff it’s hard to appreciate how far ahead Brain and DM are. Neither OpenAI nor FAIR ever has the top score on anything unless Google delays publication . And short of FAIR? D2 lacrosse. There are exceptions to such…

Not elitist at all; I highly appreciate this post. I know the basics of ML but otherwise am clueless when it comes to the true depths of this field and it's interesting to hear this perspective.

I used a lot of jargon and lingo and inside baseball in that post, it was intended for people who have deep background.

But if you’re interested I’m happy to (attempt) answers to anything that was jargon: by virtue of HN my answers will be peer-reviewed in real time, and with only modest luck, a true expert might chime in.

Re: Imagen, a text-to-image diffusion model

#335

Post by tines »

paisawalla wrote: post #310
tines wrote: post #279

Earlier quoted context omitted.

No, because there's at least one male nurse.

Please don't waste time with this kind of obtuse response. This fact says nothing about why nursing is a female-dominated career. You claim to know that this is just an accidental fact of history or society -- how do you know that?

I meant "accidental" in the Aristotelian sense: https://plato.stanford.edu/entries/essential-accidental/

Re: Imagen, a text-to-image diffusion model

#336

Post by joshcryer »

benreesman wrote: post #273

I apologize in advance for the elitist-sounding tone. In my defense the people I’m calling elite I have nothing to do with, I’m certainly not talking about myself. Without a fairly deep grounding in this stuff it’s hard to appreciate how far ahead Brain and DM are. Neither OpenAI nor FAIR ever has the top score on anything unless Google delays publication . And short of FAIR? D2 lacrosse. There are exceptions to such…

Who does it serve for plebs to be shown the approach openly? I don't know that it does a disservice to anyone by showing the approach.

But in general it is likely more due in part to the fact that it's going to happen anyway, if we can share our approaches and research findings, we'll just achieve it sooner.

Re: Imagen, a text-to-image diffusion model

#337

Post by pshc »

bufbupa wrote: post #158
karpierz wrote: post #61

Earlier quoted context omitted.

It depends on whether you'd like the model to learn casual or correlative relationships. If you want the model to understand what a "nurse" actually is, then it shouldn't be associated with female. If you want the model to understand how the word "nurse" is usually used, without regard for what a "nurse" actually is, then associating it with female is fine. The issue with a correlative model is that it can easily be…

At the end of a day, if you ask for a nurse, should the model output a male or female by default? If the input text lacks context/nuance, then the model must have some bias to infer the user's intent. This holds true for any image it generates; not just the politically sensitive ones. For example, if I ask for a picture of a person, and don't get one with pink hair, is that a shortcoming of the model? I'd say that bi…

Perhaps to avoid this issue, future versions of the model would throw an error like “bias leak: please specify a gender for the nurse at character 32”

Re: Imagen, a text-to-image diffusion model

#338

Post by thorum »

qclibre22 wrote: post #75
benwikler wrote: post #27

Would be fascinated to see the DALL-E output for the same prompts as the ones used in this paper. If you've got DALL-E access and can try a few, please put links as replies!

See the paper here : https://gweb-research-imagen.appspot.com/paper.pdf Section E : "Comparison to GLIDE and DALL-E 2"

Imagen seems better at capturing details/nuance from the prompt, but subjectively the DALLE-2 images feel more “real” to me. Not sure why. Something about the lighting?

Re: Imagen, a text-to-image diffusion model

#339

Post by magicalist »

tines wrote: post #297
magicalist wrote: post #285

Earlier quoted context omitted.

> Could skewing search results, i.e. hiding the bias of the real world Which real world? The population you sample from is going to make a big difference. Do you expect it to reflect your day to day life in your own city? Own country? The entire world? Results will vary significantly.

I'd say it doesn't actually matter, as long as the population sampled is made clear to the user. If I ask for pictures of Japanese people, I'm not shocked when all the results are of Japanese people. If I asked for "criminals in the United States" and all the results are black people, that should concern me, not because the data set is biased but because the real world is biased and we should do something about that.…

> If I asked for "criminals in the United States" and all the results are black people, that should concern me, not because the data set is biased

Well the results would unquestionably be biased. All results being black people wouldn't reflect reality at all, and hurting feelings to enact change seems like a poor justification for incorrect results.

> I'd say it doesn't actually matter, as long as the population sampled is made clear to the user.

Ok, and let's say I ask for "criminals in Cheyenne Wyoming" and it doesn't know the answer to that, should it just do its best to answer? Seem risky if people are going to get fired up about it and act on this to get "real change".

That seems like a good parallel to what we're talking about here, since it's very unlikely that crime statistics were fed into this image generating model.

Re: Imagen, a text-to-image diffusion model

#340

Post by sinenomine »

hn_throwaway_99 wrote: post #298
londons_explore wrote: post #251

Earlier quoted context omitted.

Figure A.4 in the linked paper is a good high level overview of this model. Shame it was hidden away on page 19 in the appendix! Each box you see there has a section in the paper explaining it in more detail.

Uhh, yeah, I'm going to need much more of an ELI5 than that! Looking at Figure A.4, I understand (again, at a very high-level) the first step of "Frozen Text Encoder", and I have a decent understanding of the upsampling techniques used in the last 2 diffusion model steps, but the middle "Text-to-Image Diffusion Model" step that magically outputs a 64x64 pixel image of an actual golden retriever wearing an actual blue…

> but the middle "Text-to-Image Diffusion Model" step that magically outputs a 64x64 pixel image of an actual golden retriever wearing an actual blue checkered beret and red-dotted turtleneck is where I go "WTF??".

It doesn't output it outright, it basically forms it slowly, finding and strengthening more and more finer-grained features among the dwindling noise, combining the learned associations of memorized convolutional texture primitives vs encoded text embeddings. In the limit of enough data the associations and primitives turn out composable enough to suffice for out-of-distribution benchmark scenes.

When you have a high-quality encoder of your modality into a compressed vector representation, the rest is optimization over a sufficiently high-dimensional, plastic computational substrate (model): https://moultano.wordpress.com/2020/10/18/why-deep-learning-...

It works because it should. The next question is: "What are the implications?".

Can we meaningfully represent every available modality in a single latent space, and freely interconvert composable gestalts like this https://files.catbox.moe/rmy40q.jpg ?

Post reply on HN