Dylan Martin (@DylanMartinLSC) header
Dylan Martin (@DylanMartinLSC) profile picture

Computer vision.

Oklahoma
Joined November 2023
Photo by Dylan Martin (@DylanMartinLSC)Photo by Dylan Martin (@DylanMartinLSC)Photo by Dylan Martin (@DylanMartinLSC)Photo by Dylan Martin (@DylanMartinLSC)Photo by Dylan Martin (@DylanMartinLSC)Photo by Dylan Martin (@DylanMartinLSC)Photo by Dylan Martin (@DylanMartinLSC)Photo by Dylan Martin (@DylanMartinLSC)Photo by Dylan Martin (@DylanMartinLSC)Photo by Dylan Martin (@DylanMartinLSC)
Dylan Martin retweeted
Can an image generator learn its own hierarchy of abstractions without a separately trained encoder? Meet HiT. Trained end to end from pixels, it learns to dynamically group patches across multiple levels—and semantic structure emerges without masks or region labels. Check our video below: background collapses into a few groups, the main subject gets many, and words light up the regions they refer to! It also gets 1.48 FID on ImageNet 256, uses 2.4× less training compute than JiT-H/16 at matched quality, and outperforms models with 6× more parameters on average across text-to-image benchmarks. Website, paper, and code below, followed by how it works and what emerges 🧵
8
51
15
416
42,307
Dylan Martin retweeted
PyBot: No weights. No training. A chatbot built with only simple rules and no AI, entirely written in 600k lines of Python code. PyBot matches Smollm with 360M parameters! How? A swarm of 1,000 DeepSeek Flash agents wrote the whole code in just a week. github.com/huggingface/pybot
104
352
142
2,936
671,994
Dylan Martin retweeted
I only had to wait 5 years for SDF + splats to become trendy 😅 Here's pixel-Art gaussian splats before they were cool ✌️
🍍#clayxels v1.6 is out on the @AssetStore with: a whole new way of rendering splats without polygons, faster compute clay, direct viewport picking, noise primitive, clay-groups, M1 Mac support🍎
12
9
215
6,143
Generalization is a data problem not a training problem
"Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes." Charles Goodhart, 1975
32
Dylan Martin retweeted
a 364K param system-one model can sit in the SRAM of your SSD and protect it from ransomware. (88 KB) here's how:
21
53
6
1,053
41,936
I'm working on the only image gen model in the world to have cc0 training data, weights, and code. Which is less of a brag and more of a representation on how sad the current CV industry has become.
31
Dylan Martin retweeted
3D gaussian pixel sharks
36
876
15
9,461
115,321
Literally the coolest shit I've seen this week
Trained this 9M param world model This is NOT using a rendering engine. A NEURAL NETWORK is generating each frame at 30fps conditioned directly on user input The model inputs past K frames, latest user action, and generates the new frame! It learned to respect light interactions, collisions, and long-term continuity! Done using a VQ-VAE and a Transformer. On the left you can see the VQ codewords it generated which then gets rendered onto the frame!
1
78
Dylan Martin retweeted
樹や花の3Dモデルを簡単に作れるツール「自然物ガチャ」を作成しました!! 無料で使えて、微調整も簡単です。商用利用もok。 ぜひ使ってみてください!! tools.oyasumi-gamedev.com/sh…
架空の宇宙、空を数式で表現する「空ガチャ」を公開しました!! 無料で使えて漫画のトーンにも変換できます。 様々な星空、豊富な種類の雲が流れゆくさまをお楽しみください!!登録不要で商用利用もok。 是非使ってみてください! tools.oyasumi-gamedev.com/so…
11
374
33
1,734
151,452
Dylan Martin retweeted
シミュレーションでマテリアルを作成する、新感覚のテクスチャ作成ツールを開発しました!! tools.oyasumi-gamedev.com/to… その名も「トーン・マテリアルガチャ」!! 物の成り立ちを再現するツールです!! 無料で使えて、ガチャで無限に近いバリエーションを試せます。しかもパラメーターで微調整も可能。漫画のトーンにも変換できます。 ■何がすごいのか? 従来のマテリアル作成ソフトは、図形の変形で質感を作っていました。 このソフトでは、岩は地層から、樹皮は樹の育ち方をシュミレーションして、実物と同じ仕組みでマテリアルを作成することができます。(タイルなどは開発中) 商用利用も可能です。ぜひ使ってみてください!!
樹や花の3Dモデルを簡単に作れるツール「自然物ガチャ」を作成しました!! 無料で使えて、微調整も簡単です。商用利用もok。 ぜひ使ってみてください!! tools.oyasumi-gamedev.com/sh…
17
315
18
2,494
86,744
You know it's a good chatgpt convo when the summary is in Chinese
26
If you draw pixel art, small art, or would like to fill in some gaps: please contribute to the AMBER-pixel64 project. Sprites must be hand drawn, CC0 or public domain. Image gen can only be as good as its data. github.com/DylanMartinLSC/am…
1
64
I'm releasing AMBER-64, a ~100k pixel art image dataset (64x64 resolution). It features over 75,000 handmade sprites, as well as thousands of converted vectors, photos and 3d models. To the best of my ability, I ensured no AI art entered the dataset.
2
61
AMBER stands for "All Made By Ethical Resources" as all images contained in the dataset are either CC0 or public domain, and respected artists who did not want their data to be trained on. The name is an ode to my beautiful wife, Amber.
14
Dylan Martin retweeted
Replying to @yunta_tsai
A petition is underway to reinstate Jensen and I’m signing it too
7,507
20,258
3,134
229,030
10,878,044
Dylan Martin retweeted
I have a major result to share. I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China. This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model. Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves. I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems. I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used. It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20. These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs. I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo…. There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
1,820
6,427
2,200
41,228
6,733,290
Dylan Martin retweeted
i built resnet-18 in pure assembly. 11.7m params. zero libc calls. im2col, the matmul kernel, batchnorm folding, pooling, softmax, even exp(): all arm64 asm. ~45 ms per image on a single m1 core. 8 kb of machine code. top-5 matches pytorch on all 30 test images.
13
3
64
1,714
Female CEOs right now:
🤖 Made with AI
40
Dylan Martin retweeted
China wants Jensen out, so I want him in.
1,860
3,606
519
59,366
1,848,385