DiT-3D: Exploring Plain Diffusion Transformers for 3D Shape Generation

Keyphrases

• Transformer 100%
• 3D Shape Generation 100%
• Transformer Architecture 50%
• U-Net 50%
• ShapeNet 50%
• Voxelized Point Cloud 50%
• Token 25%
• Point Cloud 25%
• ImageNet 25%
• 2D Image 25%
• Computational Cost 25%
• Art Performance 25%
• State-of-the-art Techniques 25%
• Self-attention 25%
• 2D to 3D 25%
• Chamfer Distance 25%
• Parameter Tuning 25%
• Diffusion Method 25%
• Coverage Metrics 25%
• 3D Point Cloud Generation 25%
• Model Size 25%
• Design philosophy 25%
• Denoising Process 25%
• Transformer Block 25%
• Patch Embedding 25%
• 1-nearest Neighbor 25%
• Window Attention 25%
• Positional Embedding 25%

Computer Science

• Point Cloud 100%
• U-Net 50%
• de-noising 25%
• Art Performance 25%
• Experimental Result 25%
• Computational Cost 25%

Engineering

• Point Cloud 100%
• Experimental Result 25%
• Metrics 25%
• State-of-the-Art Method 25%
• Computational Cost 25%
• Nearest Neighbor 25%