{"id":25147,"date":"2022-12-21T22:12:46","date_gmt":"2022-12-21T22:12:46","guid":{"rendered":"https:\/\/mailinvest.blog\/index.php\/2022\/12\/21\/openai-gets-to-the-pointe-with-open-source-text-to-3d-model-the-register\/"},"modified":"2022-12-21T22:12:46","modified_gmt":"2022-12-21T22:12:46","slug":"openai-gets-to-the-pointe-with-open-source-text-to-3d-model-the-register","status":"publish","type":"post","link":"https:\/\/mailinvest.blog\/index.php\/2022\/12\/21\/openai-gets-to-the-pointe-with-open-source-text-to-3d-model-the-register\/","title":{"rendered":"OpenAI gets to the Point\u2022E with open source text-to-3D model \u2022 The Register"},"content":{"rendered":"

<\/a>\n<\/p>\n

\n

OpenAI has extended the capabilities of its text-to-image software from two dimensions into three with the release of Point\u2022E, an open source project that produces 3D images from text prompts.<\/p>\n

The AI research firm has attracted considerable attention for its DALL\u2022E software, which like rival projects Stable Diffusion and Midjourney<\/a> can generate realistic or fantastical images from descriptive text.<\/p>\n

While Point\u2022E<\/a> shares the bullet point symbol used in OpenAI’s DALL\u2022E branding, it relies on a different machine learning model called GLIDE. And presently, it’s not nearly as capable. Given a text directive like “a traffic cone,” Point\u2022E produces a low-resolution point cloud \u2013 a set of points in space \u2013 that resembles a traffic cone.<\/p>\n

\"Sample<\/a><\/p>\n

Sample Point\u2022E images – Click to enlarge<\/p>\n<\/div>\n

The result is nowhere near the quality of a commercial 3D rendering in a film or video game. But it’s not supposed to be. Point clouds represent an intermediate step \u2013 once fed into an 3D application like Blender, they can be turned into textured meshes which look more like familiar 3D imagery.<\/p>\n

\"Sample<\/a><\/p>\n

Sample Point\u2022E images converted to mesh – Click to enlarge<\/p>\n<\/div>\n

“While our method still falls short of the state-of-the-art in terms of sample quality, it is one to two orders of magnitude faster to sample from, offering a practical trade-off for some use cases,” explain OpenAI researchers Alex Nichol, Heewoo Jun, Prafulla Dhariwal, Pamela Mishkin, and Mark Chen in a paper<\/a> [PDF] describing the project.<\/p>\n

The point of Point\u2022E is that it “generates point clouds efficiently” \u2013 that’s where the “E” comes from in this case. It can produce 3D models using only one to two minutes of GPU time, compared to state-of-the-art methods that require multiple GPU hours to create a finished rendering. It’s significantly faster than Google’s DreamFusion<\/a> text-to-3D model \u2013 600x<\/a> by one estimate.<\/p>\n

\n