Training-Free Building Change Detection with Example-Induced Discriminants
Organizations: University of Electronic Science and Technology of China · Southwest Jiaotong University
Abstract
Building change detection (BCD) relies on dense annotations that are costly to collect across regions. A few annotated examples offer a more economical way to specify the target, but change maps provide incomplete building supervision: they label only changed buildings and leave the building-containing date unspecified. Our key observation is that these annotations nevertheless reveal features that recognize buildings in other images, including unchanged ones. We present Linear INduced Discriminants (LIND), a training-free framework that turns regional feature contrasts into a reusable linear building classifier. Using a single frozen vision transformer, it unifies building recognition and temporal comparison in the same feature space. This approach requires neither text prompts nor an auxiliary segmentation model. Only one parameter is tuned on a single validation dataset and then kept fixed across datasets and shot settings. Extensive experiments demonstrate strong detection accuracy, cross-dataset transfer, and efficient inference, with extensions to other semantic classes and broader land-cover changes. Change annotations thus provide a compact visual specification for detection in new scenes. Code: https://github.com/lzcn/lind
Figures & tables
| Method | LEVIR-CD | WHU-CD | ||
|---|---|---|---|---|
| IoU | F1 | IoU | F1 | |
| CVA [ 30 ] | 5.41 | 10.26 | 3.71 | 7.16 |
| IRMAD [ 32 ] | 6.72 | 12.59 | 3.31 | 6.41 |
| PCA-K [ 6 ] | 5.36 | 10.18 | 6.64 | 12.46 |
| ISFA [ 47 ] | 5.36 | 10.17 | 3.82 | 7.36 |
| DSFA [ 13 ] | 7.92 | 14.67 | 5.91 | 11.15 |
| Building F1 | ||||||||
|---|---|---|---|---|---|---|---|---|
| IoU | F1 | IoU | F1 | IoU | F1 | |||
| 1 | 56.83 | 72.48 | 71.17 | 83.16 | +14.34 | +10.68 | 81.14 | 93.09 |
| 5 | 80.58 | 89.25 | 81.52 | 89.82 | +0.94 | +0.57 | 93.45 | 93.54 |
| 10 | 82.84 | 90.62 | 83.12 | 90.78 | +0.28 | +0.17 | 93.71 | 93.54 |
| Source | LEVIR-CD | WHU-CD | ||
|---|---|---|---|---|
| Method | IoU | F1 | IoU | F1 |
| OpenDPR [ 17 ] | 23.71 | 38.33 | 6.31 | 11.87 |
| AdaptOVCD [ 12 ] | 24.38 | 39.20 | 4.72 | 9.01 |
| OmniOVCD [ 53 ] | 22.67 | 36.96 | 22.26 | 36.42 |
| MemOVCD [ 24 ] | 17.11 | 29.22 | 11.20 | 20.14 |
| CoRegOVCD [ 43 ] | 18.87 | 31.74 | 3.86 | 7.44 |
| Method | WHU LEVIR | LEVIR WHU | ||
|---|---|---|---|---|
| IoU | F1 | IoU | F1 | |
| AdaptOVCD [ 12 ] | 8.26 | 15.26 | 45.66 | 62.70 |
| OmniOVCD [ 53 ] | 62.06 | 76.59 | 71.47 | 83.36 |
| CoRegOVCD [ 43 ] | 32.18 | 48.70 | 61.12 | 75.87 |
| ReA-OVCD [ 57 ] | 65.30 | 79.01 | 78.72 | 88.10 |
| OpenDPR [ 17 ] | 35.33 | 52.21 | 56.43 | 72.15 |
| Method | Test set | Full set | ||
|---|---|---|---|---|
| IoU | F1 | IoU | F1 | |
| OpenDPR [ 17 ] | 37.33 | 54.37 | 22.05 | 36.13 |
| AdaptOVCD [ 12 ] | 42.21 | 59.37 | 41.69 | 58.84 |
| OmniOVCD [ 53 ] | 32.95 | 49.57 | 36.67 | 53.66 |
| MemOVCD [ 24 ] | 33.70 | 50.41 | 25.23 | 40.30 |
| CoRegOVCD [ 43 ] | 32.72 | 49.31 | 43.27 | 60.40 |
| Method | Building | Tree | Water | Low veg. | N.v.g. Surface | Playground | Avg. | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| IoU | F1 | IoU | F1 | IoU | F1 | IoU | F1 | IoU | F1 | IoU | F1 | IoU | F1 | |
| IMC [ 26 ] | 31.74 | 48.19 | 10.89 | 19.64 | 12.34 | 21.97 | 0.40 | 0.79 | 0.00 | 0.00 | 26.93 | 42.43 | 13.72 | 22.17 |
| MCI [ 26 ] | 38.65 | 55.75 | 15.38 | 26.66 | 15.39 | 26.68 | 21.17 | 34.95 | 27.91 | 43.65 | 22.39 | 36.59 | 23.48 | 37.38 |
| UniVCD [ 58 ] | 43.20 | 60.40 | 18.90 | 31.90 | 8.20 | 15.20 | 24.90 | 39.80 | 28.00 | 43.70 | 0.00 | 0.00 | 20.53 | 31.83 |
| OmniOVCD [ 53 ] | 45.20 | 62.30 | 16.70 | 28.60 | 21.20 | 35.00 | 24.50 | 39.30 | 27.70 | 43.40 | 27.00 | 42.40 | 27.05 | 41.83 |
| OpenDPR [ 17 ] | 42.40 | 59.50 | 20.90 | 34.50 | 17.50 | 29.80 | 23.00 | 37.40 | 30.20 | 46.40 | 37.30 | 54.30 | 28.55 | 43.65 |
| LEVIR-CD | WHU-CD | |||||
|---|---|---|---|---|---|---|
| 1 | 5 | 10 | 1 | 5 | 10 | |
| Full | 78.38 | 84.82 | 85.17 | 72.48 | 89.25 | 90.62 |
| (a) | 77.69 | 83.44 | 83.69 | 72.15 | 88.76 | 90.07 |
| (b) | 72.22 | 77.93 | 77.47 | 57.41 | 84.73 | 88.60 |
| (c) | 74.95 | 82.63 | 83.11 | 70.32 | 87.55 | 89.69 |
| (d) | 78.04 | 82.82 | 82.91 | 69.57 | 80.97 | 81.36 |
| LEVIR-CD | WHU-CD | S2Looking | |||||||
|---|---|---|---|---|---|---|---|---|---|
| Shift | 0 | 1 | 2 | 0 | 1 | 2 | 0 | 1 | 2 |
| 0.17 | 0.17 | 0.27 | 0.13 | 0.16 | 1.77 | 0.46 | 0.49 | 1.05 | |
| 0.56 | 0.61 | 1.61 | 0.97 | 4.41 | 9.86 | 0.96 | 2.05 | 3.93 | |
| Backbone | LEVIR-CD | WHU-CD | ||||
|---|---|---|---|---|---|---|
| DINOv2-L | 65.26 | 74.16 | 75.02 | 62.61 | 81.20 | 83.30 |
| DINOv3-L | 78.38 | 84.82 | 85.17 | 72.48 | 89.25 | 90.62 |
| DINOv3-L † | 67.87 | 78.68 | 79.77 | 61.13 | 72.66 | 75.76 |
| † SAT493M satellite pretraining. | ||||||
| Method | FMs | Latency (s) | Memory (GiB) |
|---|---|---|---|
| AdaptOVCD [ 12 ] | 3 | 4.61 | 10.32 |
| CoRegOVCD [ 43 ] | 2 | 1.51 | 5.55 |
| OmniOVCD [ 53 ] | 1 | 0.60 | 4.01 |
| OpenDPR [ 17 ] | 2 | 0.71 | 6.39 |
| ReA-OVCD [ 57 ] | 1 | 0.93 | 3.97 |
| Ours | 1 | 0.36 | 1.51 |
Remote Sensing Change Detection
Remote sensing change detection finds and interprets differences across Earth-observation images acquired at different times.
See all 101 papers