Hugging Face's logo

Datasets:
BWGZK
/
procrop_dataset

Dataset card Files
xet
 Community
BWGZK commited on
Commit
9fe10c4
·
verified ·
1 Parent(s): 56b80e2

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +6 -5
README.md CHANGED
@@ -1,10 +1,11 @@
1
- Dataset Card: CAD + Retrieval Tables + SAM Embedding Databases
2
- Dataset Summary
 
3
 
4
  This dataset release provides supporting data and precomputed features for composition-aware image cropping research. It contains three main components:
5
 
6
- CAD Dataset: weakly annotated images generated via ControlNet-based outpainting.
7
- Precomputed Retrieval Tables: .pt files mapping each query image to its top-32 retrieved references from AVA and CGL.
8
- Pre-extracted SAM Embedding Databases: parquet files containing SAM image embeddings for efficient evaluation on GAIC, FLMS, and SACD.
9
 
10
  The dataset is designed to support retrieval-augmented aesthetic image cropping, reference-based crop recommendation, and efficient benchmark evaluation without requiring users to recompute retrieval or SAM features from scratch.
 
1
+ # Dataset Card: CAD + Retrieval Tables + SAM Embedding Databases
2
+
3
+ ## Dataset Summary
4
 
5
  This dataset release provides supporting data and precomputed features for composition-aware image cropping research. It contains three main components:
6
 
7
+ - **CAD Dataset:** Weakly annotated images generated via ControlNet-based outpainting.
8
+ - **Precomputed Retrieval Tables:** `.pt` files mapping each query image to its top-32 retrieved references from AVA and CGL.
9
+ - **Pre-extracted SAM Embedding Databases:** Parquet files containing SAM image embeddings for efficient evaluation on GAIC, FLMS, and SACD.
10
 
11
  The dataset is designed to support retrieval-augmented aesthetic image cropping, reference-based crop recommendation, and efficient benchmark evaluation without requiring users to recompute retrieval or SAM features from scratch.