chuanle commited on
Commit
b7d49dc
·
verified ·
1 Parent(s): 113ac0f

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +8 -7
README.md CHANGED
@@ -13,11 +13,11 @@ library_name: generic
13
 
14
  Official pretrained checkpoints for **ReDiCo**, a 3D Referring Expression Segmentation (3D-RES / 3D-GRES) method.
15
 
16
- [Code](https://github.com/songchuanle-1/ReDiCo) | [Paper](#)
17
 
18
  ## Model Description
19
 
20
- ReDiCo iteratively refines query-point features through relay-mediated visual updating, jointly modeling geometric relations, coverage, and diversity of the assigned points for 3D referring expression segmentation.
21
 
22
  Two checkpoints:
23
 
@@ -30,7 +30,7 @@ Two checkpoints:
30
 
31
  | Task | Dataset | mIoU | Acc@0.5 | Acc@0.25 |
32
  |---|---|---|---|---|
33
- | 3D-RES | ScanRefer | 51.7 | 57.1 | 62.7 |
34
  | 3D-GRES | Multi3DRefer | 53.7 | 54.1 | 73.8 |
35
 
36
  ## Usage
@@ -56,10 +56,11 @@ Data and pretrained backbone (`sp_unet_backbone.pth`) download links are in the
56
  ## Citation
57
 
58
  ```bibtex
59
- @article{redico,
60
  title = {ReDiCo: Relay-Mediated Visual Updating for 3D Referring Segmentation},
61
- author = {},
62
- journal = {},
63
- year = {2026}
 
64
  }
65
  ```
 
13
 
14
  Official pretrained checkpoints for **ReDiCo**, a 3D Referring Expression Segmentation (3D-RES / 3D-GRES) method.
15
 
16
+ [Code](https://github.com/songchuanle-1/ReDiCo) | [Paper (under review)]()
17
 
18
  ## Model Description
19
 
20
+ 3D Referring Expression Segmentation (3D-RES) aims to produce point-level masks for objects described by natural language in complex 3D scenes. **ReDiCo** (**Re**lay attention with **Di**versity and **Co**verage regularization) reformulates visual token updating as *token-relay-token* communication: a compact set of language-guided relay tokens aggregates global scene information and distributes the refined context back to the visual stream, reducing the dominant interaction complexity from $O(N_{sp}^2)$ to $O(N_q N_{sp})$ with $N_q \ll N_{sp}$. Geometry-aware relay attention and coverage/diversity regularization keep this relay bottleneck reliable.
21
 
22
  Two checkpoints:
23
 
 
30
 
31
  | Task | Dataset | mIoU | Acc@0.5 | Acc@0.25 |
32
  |---|---|---|---|---|
33
+ | 3D-RES | ScanRefer | 51.7 | 57.1 | 62.8 |
34
  | 3D-GRES | Multi3DRefer | 53.7 | 54.1 | 73.8 |
35
 
36
  ## Usage
 
56
  ## Citation
57
 
58
  ```bibtex
59
+ @article{song2026redico,
60
  title = {ReDiCo: Relay-Mediated Visual Updating for 3D Referring Segmentation},
61
+ author = {Song, Chuanle and Li, Xinyu and Lin, Shuxin and Huang, Yihua},
62
+ journal = {IEEE Transactions on Multimedia},
63
+ year = {2026},
64
+ note = {under review}
65
  }
66
  ```