Oh yeah go ahead and upsample bioshock infinite, that was everyone’s main problem with it
October 14, 2024 at 11:34 AM
Oh yeah go ahead and upsample bioshock infinite, that was everyone’s main problem with it
yeah definitely looks like finetuning, i am 50/50 between the answering as elon being undisclosed fuckery with the prompt or their finetuning dataset just being Like That. just upsample elon twenty times
July 8, 2025 at 7:45 PM
yeah definitely looks like finetuning, i am 50/50 between the answering as elon being undisclosed fuckery with the prompt or their finetuning dataset just being Like That. just upsample elon twenty times
that's insane, i think you are objectively more famous than either me or theophite. maybe they upsample tech twitter???
June 12, 2026 at 7:01 AM
that's insane, i think you are objectively more famous than either me or theophite. maybe they upsample tech twitter???
Super excited to introduce
✨ AnyUp: Universal Feature Upsampling 🔎
Upsample any feature - really any feature - with the same upsampler, no need for cumbersome retraining.
SOTA feature upsampling results while being feature-agnostic at inference time.
🌐 wimmerth.github.io/anyup/
✨ AnyUp: Universal Feature Upsampling 🔎
Upsample any feature - really any feature - with the same upsampler, no need for cumbersome retraining.
SOTA feature upsampling results while being feature-agnostic at inference time.
🌐 wimmerth.github.io/anyup/
October 16, 2025 at 9:07 AM
Super excited to introduce
✨ AnyUp: Universal Feature Upsampling 🔎
Upsample any feature - really any feature - with the same upsampler, no need for cumbersome retraining.
SOTA feature upsampling results while being feature-agnostic at inference time.
🌐 wimmerth.github.io/anyup/
✨ AnyUp: Universal Feature Upsampling 🔎
Upsample any feature - really any feature - with the same upsampler, no need for cumbersome retraining.
SOTA feature upsampling results while being feature-agnostic at inference time.
🌐 wimmerth.github.io/anyup/
Shudder you absolute mad lads
May 8, 2025 at 10:50 PM
Shudder you absolute mad lads
Roblox Reality promises to upsample blocky Lego-like Roblox games like Grow a Garden until they're photorealistic.
#pcgamer #pcgaming #roblox #robloxreality
#pcgamer #pcgaming #roblox #robloxreality
May 12, 2026 at 3:30 AM
Roblox Reality promises to upsample blocky Lego-like Roblox games like Grow a Garden until they're photorealistic.
#pcgamer #pcgaming #roblox #robloxreality
#pcgamer #pcgaming #roblox #robloxreality
Added DLSS Ray Reconstruction to Bevy Solari!
It's kind of fun to look at the profile, I _think_ this is a series of upsample/downsample convolutions for a U-net, but who knows!
It's kind of fun to look at the profile, I _think_ this is a series of upsample/downsample convolutions for a U-net, but who knows!
August 9, 2025 at 9:35 PM
Added DLSS Ray Reconstruction to Bevy Solari!
It's kind of fun to look at the profile, I _think_ this is a series of upsample/downsample convolutions for a U-net, but who knows!
It's kind of fun to look at the profile, I _think_ this is a series of upsample/downsample convolutions for a U-net, but who knows!
@wimmerthomas.bsky.social introduces AnyUp: Universal Feature Upsampling
Upsample any feature - really any feature - with the same upsampler, no need for cumbersome retraining while being feature-agnostic at inference time.
Upsample any feature - really any feature - with the same upsampler, no need for cumbersome retraining while being feature-agnostic at inference time.
October 16, 2025 at 5:13 AM
@wimmerthomas.bsky.social introduces AnyUp: Universal Feature Upsampling
Upsample any feature - really any feature - with the same upsampler, no need for cumbersome retraining while being feature-agnostic at inference time.
Upsample any feature - really any feature - with the same upsampler, no need for cumbersome retraining while being feature-agnostic at inference time.
oops- i forgot to upsample it so it wouldn't be blurred by bsky's compression #webbborder1
May 24, 2026 at 6:53 PM
oops- i forgot to upsample it so it wouldn't be blurred by bsky's compression #webbborder1
I think this is precisely the point. Not all LLMs are the same just because they’re LLMs. People make choices about training—what data to include, what the mix should be, whether to upsample, and a million other little things. They don’t just happen, and they aren’t foregone conclusions.
2️⃣ a mistake in my early analysis of fair use in LLMs was comparing Google Book Search (a single product with specific knowable characteristics) with the entire class of LLMs. Experiments remind us that we have to be careful when generalizing, since models may have meaningfully different properties.
May 21, 2025 at 8:24 PM
I think this is precisely the point. Not all LLMs are the same just because they’re LLMs. People make choices about training—what data to include, what the mix should be, whether to upsample, and a million other little things. They don’t just happen, and they aren’t foregone conclusions.
you simply need to find a large and unbiased dataset of non-yassified women in very high resolution to downsample and then train the model to upsample, how hard could it be
March 18, 2026 at 10:39 PM
you simply need to find a large and unbiased dataset of non-yassified women in very high resolution to downsample and then train the model to upsample, how hard could it be
“Hey, can you upsample and color correct this photo of the placenta from when my daughter was born? I need it for an art project.”
- Random Things Friends Ask Me
(The photo was gnarly, amazing, and not for the squeamish BTW)
- Random Things Friends Ask Me
(The photo was gnarly, amazing, and not for the squeamish BTW)
January 30, 2025 at 7:58 PM
“Hey, can you upsample and color correct this photo of the placenta from when my daughter was born? I need it for an art project.”
- Random Things Friends Ask Me
(The photo was gnarly, amazing, and not for the squeamish BTW)
- Random Things Friends Ask Me
(The photo was gnarly, amazing, and not for the squeamish BTW)
Do you need to upsample vision features (e.g. DINOv3) to higher resolutions?
👉 If yes, check out Thomas Wimmer's new work AnyUp, it works on all vision features without retraining and can upsample to any resolution! Code is available!
👉 If yes, check out Thomas Wimmer's new work AnyUp, it works on all vision features without retraining and can upsample to any resolution! Code is available!
Super excited to introduce
✨ AnyUp: Universal Feature Upsampling 🔎
Upsample any feature - really any feature - with the same upsampler, no need for cumbersome retraining.
SOTA feature upsampling results while being feature-agnostic at inference time.
🌐 wimmerth.github.io/anyup/
✨ AnyUp: Universal Feature Upsampling 🔎
Upsample any feature - really any feature - with the same upsampler, no need for cumbersome retraining.
SOTA feature upsampling results while being feature-agnostic at inference time.
🌐 wimmerth.github.io/anyup/
October 16, 2025 at 11:51 AM
Do you need to upsample vision features (e.g. DINOv3) to higher resolutions?
👉 If yes, check out Thomas Wimmer's new work AnyUp, it works on all vision features without retraining and can upsample to any resolution! Code is available!
👉 If yes, check out Thomas Wimmer's new work AnyUp, it works on all vision features without retraining and can upsample to any resolution! Code is available!
I think it's just that polls which are more favorable to both are ones that upweight or upsample old people
March 2, 2026 at 12:07 AM
I think it's just that polls which are more favorable to both are ones that upweight or upsample old people
Not sure how to show it, but I implementet a bilateral upsampler for my fog and cloud buffer.
It helps my downsampled low frequency translucent stuff to upsample into the full res backbuffer and still keep edges on things in front- and intersecting nice and clean.
#gamedev #indiedev #vulkan
It helps my downsampled low frequency translucent stuff to upsample into the full res backbuffer and still keep edges on things in front- and intersecting nice and clean.
#gamedev #indiedev #vulkan
August 31, 2026 at 11:53 AM
It probably has to upsample it from infinity to 384kbps first.
May 25, 2026 at 6:14 PM
It probably has to upsample it from infinity to 384kbps first.
zoom in on 4 pixels and upsample to full HD 4K?
April 17, 2025 at 5:27 PM
zoom in on 4 pixels and upsample to full HD 4K?
Handheld mode is only 1080p?
No way is it actually rendering full native 4K on tv then, they're just saying it'll upsample to 4k.
No way is it actually rendering full native 4K on tv then, they're just saying it'll upsample to 4k.
April 3, 2025 at 4:40 PM
Handheld mode is only 1080p?
No way is it actually rendering full native 4K on tv then, they're just saying it'll upsample to 4k.
No way is it actually rendering full native 4K on tv then, they're just saying it'll upsample to 4k.
#readandplay (313)
#MusicChallenge 🎶🎧
“No rules. The only limit is our imagination.”
One, two, three, four, five, six, seven, eight. Schlemiel! Schlimazel! Hasenpfeffer Incorporated! 😀
"Making Our Dreams Come True" by Cyndi Grecco
youtu.be/7end5IlqD8k?...
#MusicChallenge 🎶🎧
“No rules. The only limit is our imagination.”
One, two, three, four, five, six, seven, eight. Schlemiel! Schlimazel! Hasenpfeffer Incorporated! 😀
"Making Our Dreams Come True" by Cyndi Grecco
youtu.be/7end5IlqD8k?...
April 10, 2025 at 5:04 PM
#readandplay (313)
#MusicChallenge 🎶🎧
“No rules. The only limit is our imagination.”
One, two, three, four, five, six, seven, eight. Schlemiel! Schlimazel! Hasenpfeffer Incorporated! 😀
"Making Our Dreams Come True" by Cyndi Grecco
youtu.be/7end5IlqD8k?...
#MusicChallenge 🎶🎧
“No rules. The only limit is our imagination.”
One, two, three, four, five, six, seven, eight. Schlemiel! Schlimazel! Hasenpfeffer Incorporated! 😀
"Making Our Dreams Come True" by Cyndi Grecco
youtu.be/7end5IlqD8k?...
my headcanon is that those are the same people who upsample anime to 60fps and think that it looks good
February 28, 2024 at 7:04 PM
my headcanon is that those are the same people who upsample anime to 60fps and think that it looks good
I mostly thought that one would upsample the prompt w/ a LLM before passing too LTV - hence the level of detail.
Probably also the curse of going from a low dimensional (text) representation to high dimensional too.
Probably also the curse of going from a low dimensional (text) representation to high dimensional too.
November 23, 2024 at 11:04 PM
I mostly thought that one would upsample the prompt w/ a LLM before passing too LTV - hence the level of detail.
Probably also the curse of going from a low dimensional (text) representation to high dimensional too.
Probably also the curse of going from a low dimensional (text) representation to high dimensional too.
This is what we do at Guerrilla:
- Bilinear upsample the mip to original res
- Find the bin in the histogram of the upsampled mip where the coverage matches the original
- Scale the mip's values to move that bin to whatever cutoff you use
Works with regular alpha and SDF.
- Bilinear upsample the mip to original res
- Find the bin in the histogram of the upsampled mip where the coverage matches the original
- Scale the mip's values to move that bin to whatever cutoff you use
Works with regular alpha and SDF.
January 4, 2025 at 6:19 PM
This is what we do at Guerrilla:
- Bilinear upsample the mip to original res
- Find the bin in the histogram of the upsampled mip where the coverage matches the original
- Scale the mip's values to move that bin to whatever cutoff you use
Works with regular alpha and SDF.
- Bilinear upsample the mip to original res
- Find the bin in the histogram of the upsampled mip where the coverage matches the original
- Scale the mip's values to move that bin to whatever cutoff you use
Works with regular alpha and SDF.
Chambon et al., "NAF: Zero-Shot Feature Upsampling via Neighborhood Attention Filtering"
Essentially learning create an upsample kernel via looking at encoded images. Generalizes well to various foundational models. + it beats bicubic upsampling.
Essentially learning create an upsample kernel via looking at encoded images. Generalizes well to various foundational models. + it beats bicubic upsampling.
November 25, 2025 at 9:27 PM
Chambon et al., "NAF: Zero-Shot Feature Upsampling via Neighborhood Attention Filtering"
Essentially learning create an upsample kernel via looking at encoded images. Generalizes well to various foundational models. + it beats bicubic upsampling.
Essentially learning create an upsample kernel via looking at encoded images. Generalizes well to various foundational models. + it beats bicubic upsampling.