Datasets:
Move metadata.jsonl into images/ so ImageFolder picks up metadata
Browse filesImageFolder only loads metadata.jsonl from within the data_files
directory tree. With metadata.jsonl at root and data_files pointing
to images/**, the metadata was ignored (all columns null except image).
Now metadata.jsonl lives alongside the images in images/, with
file_name paths relative to that directory (e.g. "NEET_2024_T3/...").
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- .gitattributes +0 -1
- README.md +1 -1
- configs/benchmark_config.yaml +2 -2
- metadata.jsonl → images/metadata.jsonl +2 -2
- src/benchmark_runner.py +3 -4
.gitattributes
CHANGED
|
@@ -1,4 +1,3 @@
|
|
| 1 |
images/** filter=lfs diff=lfs merge=lfs -text
|
| 2 |
-
metadata.jsonl filter=lfs diff=lfs merge=lfs -text
|
| 3 |
images/**/*.png filter=lfs diff=lfs merge=lfs -text
|
| 4 |
*.tar.gz filter=lfs diff=lfs merge=lfs -text
|
|
|
|
| 1 |
images/** filter=lfs diff=lfs merge=lfs -text
|
|
|
|
| 2 |
images/**/*.png filter=lfs diff=lfs merge=lfs -text
|
| 3 |
*.tar.gz filter=lfs diff=lfs merge=lfs -text
|
README.md
CHANGED
|
@@ -19,7 +19,7 @@ configs:
|
|
| 19 |
- config_name: default
|
| 20 |
data_files:
|
| 21 |
- split: test
|
| 22 |
-
path: "images/**
|
| 23 |
drop_labels: true
|
| 24 |
dataset_info:
|
| 25 |
features:
|
|
|
|
| 19 |
- config_name: default
|
| 20 |
data_files:
|
| 21 |
- split: test
|
| 22 |
+
path: "images/**"
|
| 23 |
drop_labels: true
|
| 24 |
dataset_info:
|
| 25 |
features:
|
configs/benchmark_config.yaml
CHANGED
|
@@ -15,8 +15,8 @@ openrouter_models:
|
|
| 15 |
# - "anthropic/claude-3-haiku"
|
| 16 |
|
| 17 |
# Path to the metadata JSONL file and base directory for image paths.
|
| 18 |
-
metadata_path: "metadata.jsonl"
|
| 19 |
-
images_base_dir: "
|
| 20 |
|
| 21 |
# Base directory where results for each model will be saved.
|
| 22 |
results_base_dir: "results"
|
|
|
|
| 15 |
# - "anthropic/claude-3-haiku"
|
| 16 |
|
| 17 |
# Path to the metadata JSONL file and base directory for image paths.
|
| 18 |
+
metadata_path: "images/metadata.jsonl"
|
| 19 |
+
images_base_dir: "images"
|
| 20 |
|
| 21 |
# Base directory where results for each model will be saved.
|
| 22 |
results_base_dir: "results"
|
metadata.jsonl → images/metadata.jsonl
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a3cc68977d9f971da43d1fae98572a0600f2b9e67b8d815a22f15dce500202fd
|
| 3 |
+
size 128832
|
src/benchmark_runner.py
CHANGED
|
@@ -476,8 +476,8 @@ def run_benchmark(
|
|
| 476 |
os.makedirs(base_output_dir, exist_ok=True)
|
| 477 |
|
| 478 |
# Load dataset directly from metadata.jsonl and images/
|
| 479 |
-
metadata_path = config.get("metadata_path", "metadata.jsonl")
|
| 480 |
-
images_base_dir = config.get("images_base_dir", "
|
| 481 |
try:
|
| 482 |
records = []
|
| 483 |
with open(metadata_path, 'r', encoding='utf-8') as f:
|
|
@@ -734,9 +734,8 @@ def run_benchmark(
|
|
| 734 |
if __name__ == "__main__":
|
| 735 |
# Get available choices for arguments
|
| 736 |
# Assuming benchmark_config.yaml is in a 'configs' directory relative to script or a fixed path
|
| 737 |
-
# Assuming metadata.jsonl is in a 'data' directory relative to script or a fixed path
|
| 738 |
default_config_path = "configs/benchmark_config.yaml"
|
| 739 |
-
default_metadata_path = "metadata.jsonl"
|
| 740 |
|
| 741 |
available_models = get_available_models(default_config_path)
|
| 742 |
available_exam_names, available_exam_years = get_available_exam_details(default_metadata_path)
|
|
|
|
| 476 |
os.makedirs(base_output_dir, exist_ok=True)
|
| 477 |
|
| 478 |
# Load dataset directly from metadata.jsonl and images/
|
| 479 |
+
metadata_path = config.get("metadata_path", "images/metadata.jsonl")
|
| 480 |
+
images_base_dir = config.get("images_base_dir", "images")
|
| 481 |
try:
|
| 482 |
records = []
|
| 483 |
with open(metadata_path, 'r', encoding='utf-8') as f:
|
|
|
|
| 734 |
if __name__ == "__main__":
|
| 735 |
# Get available choices for arguments
|
| 736 |
# Assuming benchmark_config.yaml is in a 'configs' directory relative to script or a fixed path
|
|
|
|
| 737 |
default_config_path = "configs/benchmark_config.yaml"
|
| 738 |
+
default_metadata_path = "images/metadata.jsonl"
|
| 739 |
|
| 740 |
available_models = get_available_models(default_config_path)
|
| 741 |
available_exam_names, available_exam_years = get_available_exam_details(default_metadata_path)
|