Dacheng Tao, Lei Chen, Yong Luo, Hao Yan, Bo Du, Lefei Zhang, Jingfeng Yao, Zengmao Wang, Ruiheng Zhang, Xiao He, Huangxuan Zhao, Zhou Wei
We lifted 10 functions out of this paper's own repositories and ran 6 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.
| Repository | Role | Ran |
|---|---|---|
| ZrH42/UniX | canonical | 6 of 10 |
| Function | Status | Where it lives |
|---|---|---|
| apply_rotary_pos_emb | Ran | ZrH42/UniX/modeling/unix/siglip_navit.py code served (permissive licence) · get_code("4102c144078044c7") |
| expand2square | Ran | ZrH42/UniX/modeling/unix_vlm/models/image_processing_vlm.py code served (permissive licence) · get_code("592b3c1a88f93d7c") |
| get_1d_sincos_pos_embed_from_grid | Ran | ZrH42/UniX/modeling/unix/modeling_utils.py code served (permissive licence) · get_code("e5947aba1d10885f") |
| get_2d_sincos_pos_embed | Ran | ZrH42/UniX/modeling/unix/modeling_utils.py code served (permissive licence) · get_code("58f584dfd7b8fc3f") |
| pad_sequence | Ran | ZrH42/UniX/modeling/unix/qwen2_navit.py code served (permissive licence) · get_code("26da3ed73c66b411") |
| rotate_half | Ran | ZrH42/UniX/modeling/unix/siglip_navit.py code served (permissive licence) · get_code("cec833a99feb528c") |
| create_repa_components | Not yet run | ZrH42/UniX/modeling/unix/repa_utils.py code served (permissive licence) · get_code("1fa58c6502f6cb02") |
| create_siglip_vit | Not yet run | ZrH42/UniX/modeling/unix_vlm/models/siglip_vit.py code served (permissive licence) · get_code("fa9c707483e81466") |
| get_2d_sincos_pos_embed_from_grid | Not yet run | ZrH42/UniX/modeling/unix/modeling_utils.py code served (permissive licence) · get_code("665d8a4e8f673a4c") |
| get_conv_template | Not yet run | ZrH42/UniX/modeling/unix_vlm/utils/conversation.py code served (permissive licence) · get_code("6af454338dc8d2a5") |
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
Despite recent progress, medical foundation models still struggle to unify visual understanding and generation, as these tasks have inherently conflicting goals: semantic abstraction versus pixel-level reconstruction. Existing approaches, typically based on parameter-shared autoregressive architectures, frequently lead to compromised performance in one or both tasks. To address this, we present UniX, a next-generation unified medical foundation model for chest X-ray understanding and generation. UniX decouples the two tasks into an autoregressive branch for understanding and a diffusion branch for highfidelity generation. Crucially, a cross-modal self-attention mechanism is introduced to dynamically guide the generation process with understanding features. Coupled with a rigorous data cleaning pipeline and a multi-stage training strategy, this architecture enables synergistic collaboration between tasks while leveraging the strengths of diffusion models for superior generation. On two representative benchmarks, UniX achieves a 46.1% improvement in understanding performance (Micro-F1) and a 24.2% gain in generation quality (FD-RadDino), using only a quarter of the parameters of LLM-CXR. By achieving performance on par with task-specific models, our work establishes a scalable paradigm for synergistic medical image understanding and generation. Codes and models are available at https://github.com/ZrH42/UniX.
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2601.11522")
get_code_for_paper("2601.11522")
have("2601.11522")
Connect an agent — have() is free.