SikongJueluo/Mini-Nav
1
0
Fork 0
mirror of https://github.com/SikongJueluo/Mini-Nav.git synced 2026-07-12 20:15:31 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat: vectorize CAM retrieval with NumPy and add multi-worker support

Browse Source 
- Replace scalar hamming distance with NumPy bitwise_count for batch retrieval
- Add ThreadPoolExecutor-based multi-worker query parallelism
- Improve missing dataset error message with generation command hint
- Increase DEFAULT_MAX_QUERIES from 128 to 8192 for meaningful throughput tests
This commit is contained in:
SikongJueluo
2026-06-04 16:57:53 +08:00
parent e5b764520c
commit b5a40819cc
5 changed files with 283 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
scripts/prepare_cam_retrieval_dataset.py
View File

@@ -24,6 +24,7 @@ from utils import get_device # noqa: E402
DEFAULT_COMPRESSOR_PATH = Path("outputs/hash_compressor.pt")
DEFAULT_OUTPUT_ROOT = Path("outputs/cam_retrieval_benchmark/datasets")
DEFAULT_MAX_QUERIES = 8192
@dataclass(frozen=True)
@@ -207,7 +208,7 @@ def prepare_artifact(
def main(
dataset: Literal["cifar10", "cifar100"] = typer.Option("cifar100"),
num_rows: int = typer.Option(512, min=5),
max_queries: int = typer.Option(128, min=1),
max_queries: int = typer.Option(DEFAULT_MAX_QUERIES, min=1),
compressor_path: Path = typer.Option(DEFAULT_COMPRESSOR_PATH),
output_root: Path = typer.Option(DEFAULT_OUTPUT_ROOT),
dino_model: str = typer.Option("facebook/dinov2-large"),