[HN Gopher] What If We Recaption Billions of Web Images with LLa...
       ___________________________________________________________________
        
       What If We Recaption Billions of Web Images with LLaMA-3?
        
       Author : Jimmc414
       Score  : 77 points
       Date   : 2024-06-13 03:44 UTC (19 hours ago)
        
 (HTM) web link (arxiv.org)
 (TXT) w3m dump (arxiv.org)
        
       | Jimmc414 wrote:
       | Code: https://github.com/UCSC-VLAA/Recap-DataComp-1B
        
       | interloxia wrote:
       | I'm a bit surprised by the examples. The original captions
       | contain extra context rather than descriptions of the images.
       | 
       | Perhaps that doesn't matter, as the abstract suggests that's not
       | the purpose.
       | 
       | On the other hand, training on captions they generate that are
       | also incorrect is problematic. Perhaps they should have cherry-
       | picked better examples without errors, such as details about the
       | sugar and cream, reflections of the train in the water, or the
       | location of the watermark and sign. I didn't look at the code,
       | but perhaps each image feature can be written with a confidence
       | score.
        
         | petercooper wrote:
         | Yes, the "Western Kingbird" is a good example of that. That's
         | very useful context. I've not read the paper yet, but this
         | could be trivially solved by just appending the extended
         | description to the original one. You could ask the model if the
         | original description seems accurate or not as well to weed out
         | any duds (like the picture of a cake described as a 'twin
         | room').
        
         | skerit wrote:
         | > The original captions contain extra context rather than
         | descriptions of the images
         | 
         | Another good example on how accessibility can even help people
         | without disabilities. You can see this on ALT text on Mastodon,
         | and I find it really useful.
        
       | nottorp wrote:
       | Aren't the large public models censored? Who's going to caption
       | the "offensive" images? And the porn?
        
         | lwansbrough wrote:
         | Well apparently you can just ablate those models!
         | https://news.ycombinator.com/item?id=40665721
        
           | GaggiX wrote:
           | You cannot ablate the lack of knowledge, the vision models
           | and projection layers used in this project usually lack any
           | knowledge about NSFW, the model is simply unaware of what
           | it's going on if you give them a NSFW image.
        
             | nottorp wrote:
             | Why assume offensive means NSFW?
             | 
             | I'm saying these models are censored for the lowest common
             | denominator.
        
               | robertlagrant wrote:
               | You mentioned pornography as well. And when people use
               | "offensive" in an absolute sense, that often is
               | synonymous with "what the HR department wouldn't want you
               | looking at on a work computer", i.e. NSFW.
        
               | nottorp wrote:
               | I said offensive and porn. Separately. For a reason.
        
               | robertlagrant wrote:
               | Yes, you mentioned pornography. If you don't think that's
               | safe for work, then why are you complaining about people
               | talking about NSFW material?
        
               | GaggiX wrote:
               | I didn't assume anything, you mentioned porn.
        
       | grondilu wrote:
       | I'm quite baffled by the fact that LLMs can generate a dataset
       | used to train other LLMs. One would think that such a feedback
       | loop would produce utter nonsense but apparently not. This seems
       | to work.
        
         | cjg wrote:
         | Humans have bootstrapped by training the next generation. Why
         | not LLMs?
        
           | spiderfarmer wrote:
           | I think the perception is that humans can discover new
           | information to question and improve what they learned, while
           | LLM's cannot.
        
             | immibis wrote:
             | Human language drifts for the same reason LLM language
             | would, but is continually reset to _a_ sensible state by
             | interaction with the real world.
        
         | DebtDeflation wrote:
         | Yes, and there's even a name for it and associated area of
         | research.
         | 
         | https://en.wikipedia.org/wiki/Model_collapse
        
         | bilater wrote:
         | I feel the same way about synthetic data. Seems intuitively
         | wrong that you can get new insights / unlock new abilities from
         | generated data that you could not from the original data.
        
         | throwthrowuknow wrote:
         | If the correct labels in the original training set outweigh the
         | incorrect ones then it is possible to reduce the number of
         | errors by relabeling using the trained model. If you can also
         | identify labels that are likely to be incorrect and then have
         | humans focus on relabeling those you have a way to efficiently
         | improve the data.
        
       | michaelt wrote:
       | _> first, we fine-tune a LLaMA-3-8B powered LLaVA-1.5 and then
       | employ it to recaption ~1.3 billion images from the DataComp-1B
       | dataset_
       | 
       | It's a bit surprising this works as well as they report.
       | 
       | After all, you would think if you trained LLaVA on a dataset with
       | low quality labels, you'd get a model that could only generate
       | low quality labels, and re-labelling the dataset with the model
       | would give you labels no better than the original low quality
       | ones.
        
         | mjburgess wrote:
         | Thew LLaMA model hasnt been trained on the low quality labels.
        
         | Felomo wrote:
         | If you train it on a belcurve, you will be able to fix the left
         | side with it after.
        
       | moconnor wrote:
       | I'm not convinced the paper shows this "works". Table 3 clearly
       | shows zero-shot image classification is worse with the
       | recaptioned labels.
       | 
       | All the results that show improvement seem to be evaluations
       | using LLMs, e.g. they are showing LLMs think the LLM-generated
       | text is better, which is neither surprising nor expected to
       | correlate with real downstream task performance - unless your
       | final task is labelling for an LLM, e.g. retrieval I guess.
        
         | 42lux wrote:
         | Yeah, I don't understand why they wouldn't use the captions
         | that are already present as inputs as well.
        
           | estebank wrote:
           | That's what you'd do if you wanted to try and improve the
           | captions, not if you wanted to evaluate the LLM's quality.
        
       | vessenes wrote:
       | Some good some bad here; it looks like you can in general get
       | better descriptions out of LLama-3+Llava-1.5 than random crappy /
       | short / possibly wrong human captions.
       | 
       | The image generation models they train using the new caption
       | system have better coherence and look a lot better / less weird.
       | (I don't know if they showed training detail differences between
       | their comparisons, but let's assume they're not hiding the ball
       | here).
       | 
       | The bad: Those captions don't always perform better on all tests
       | using existing infra - some of their tests perform worse against
       | existing benchmarks.
       | 
       | Why they perform worse is an open question; the answer might be
       | that the existing test data expects short / human answers, or it
       | might be that the models are more verbose but less accurate
       | (although that seems unlikely looking at the examples they show).
       | 
       | Also troubling to me is a toss-off note in Table 3 showing that
       | the "concat" data, one in which they use original caption
       | concatenated to the generated caption, has really bad CLIP
       | outcomes. That ... surprises me based on the rest of the results.
       | 
       | That said, their baseline comparison numbers in Table 1 are
       | pretty compelling -- this would definitely go into the 'try it'
       | category based on the paper. But, it also goes into the 'might
       | fail in really surprising ways, so proceed carefully' category
       | for me.
       | 
       | EDIT: Another way to think of this model is to say "can we
       | exfiltrate multi-model training data from open weights / closed
       | training models?" And I think there the answer is yes.
        
         | triyambakam wrote:
         | > EDIT: Another way to think of this model is to say "can we
         | exfiltrate multi-model training data from open weights / closed
         | training models?" And I think there the answer is yes.
         | 
         | Could you elaborate on this? Do you mean to say that this work
         | reveals some of LLaMa's training data, or that this can be used
         | to generate training data for new models?
        
           | vessenes wrote:
           | Since they're using LLama with Llava, they're getting lots of
           | Llama's baked-in understanding of the world. And I believe
           | that Llama had multimodal training data as part of its
           | pipeline, whether or not we get access to the heads that
           | parse that data in the open weights.
           | 
           | So, I'm saying that Llama has some ability to describe /
           | complete captions of photos, and very probably that ability
           | comes from some multimodal training done at FB, and that
           | using it in this way to train something new that beats
           | benchmarks is a form of pulling out the value of that
           | training data.
           | 
           | p.s. In the original comment, I meant to type 'multimodal'
           | and didn't, or got autocorrected and didn't notice -
           | apologies
        
       | bastien2 wrote:
       | Only if you have actual humans reviewing the output while being
       | paid livable wages (i.e., you don't get to use digital sweatshops
       | for this).
       | 
       | But considering Google has had over a decade to get autogenerated
       | captioning right and they consistently fail at it, I doubt any
       | slopshop is going to produce worthwhile output.
        
         | brianbreslin wrote:
         | Not to start a political argument, but I am genuinely curious
         | what you would consider a livable wage for a person in say
         | Bangladesh doing this work?
        
           | Jabrov wrote:
           | About $1000 a month would be livable in Bangladesh
        
             | kgwgk wrote:
             | Definitely "livable" considering that the average household
             | income per capita is an order of magnitude lower.
        
             | pmontra wrote:
             | 1000 Euro a month is something many young people don't get
             | paid in Western Europe.
        
       | greenhearth wrote:
       | And what if we don't?
        
       | galaxyLogic wrote:
       | The caption is like the title of a painting. It does not so much
       | tell what the image is an image of, but what the author/painter
       | tells us their image represents. Represents, not depicts.
       | 
       | Or consider a song, it's title does not tell where the sound was
       | recorded, but what the composer tells us the song is about. Not
       | where and when it was recorded.
        
       | salamo wrote:
       | I guess it depends on the purpose. As a query rewrite, these
       | cover much more of the physical scene, but lack accuracy and
       | specificity.
       | 
       | A couple of issues. First, the model completely botches the
       | "Trike" brotherhood to "Tike", which also drops the meaning of
       | the symbol. It's a tricycle. The seat isn't a sidecar, it's
       | behind the driver's. Some of the numbers for the racecars also
       | seem to be hallucinated.
       | 
       | It also doesn't seem accurate to describe the rewritten captions
       | as having a "richer" vocabulary as the authors do. It's certainly
       | more verbose and captures a lot of details about the setting, but
       | lacks information density for the subject of the photo. You can't
       | make up for low specificity with more high-probability words.
       | Sometimes you need a long-tail word.
       | 
       | For example, the Western Kingbird caption. It's not just a grey
       | and yellow bird. If the model doesn't know what a Western
       | Kingbird is, you're not going to be able to draw a Western
       | Kingbird no matter how verbose your description is.
        
         | crote wrote:
         | It's the difference between a caption and a description. Both
         | are useful, just in different settings.
         | 
         | A visually-impaired person reading the newspaper just wants to
         | know that there is an image of, say, George Washington and John
         | Adams. What do they look like? What are they wearing? What is
         | the setting? Probably not important, it's just decoration for
         | the article.
         | 
         | On the other hand, a visually-impaired person who's shopping
         | for clothing might come across a t-shirt with a print and want
         | to know what's on there. What colors are used? Is there any
         | text? Are there logos? Stuff like that matters when you're
         | choosing clothing because you don't want to end up looking like
         | a clown.
        
       ___________________________________________________________________
       (page generated 2024-06-13 23:02 UTC)