Update README.md

README.md

@@ -1,149 +0,0 @@
-
----
-license: cc-by-4.0 # Or your actual license
-tags:
-  - image-object-detection
-  - fmiyc
-  - out-of-distribution-detection
-pretty_name: "FindMeIfYouCan (FMIYC) Dataset"
-dataset_info:
-  features:
-    - name: file_name
-      dtype: string
-    - name: image_id
-      dtype: string
-    - name: height
-      dtype: float32 # Script outputs float
-    - name: width
-      dtype: float32  # Script outputs float
-    - name: dataset_origin
-      dtype: string
-    - name: distance_category
-      dtype: string
-    - name: objects
-      sequence:     # 'objects' is a list of dictionaries
-        # Fields within each object dictionary:
-        - name: id
-          dtype: string
-        - name: area
-          dtype: float32
-        - name: bbox_x      # FLATTENED
-          dtype: float32
-        - name: bbox_y      # FLATTENED
-          dtype: float32
-        - name: bbox_width  # FLATTENED
-          dtype: float32
-        - name: bbox_height # FLATTENED
-          dtype: float32
-        - name: category_id
-          dtype: int64
-    - name: categories
-      sequence: # 'categories' is a list of dictionaries
-        - name: id
-          dtype: int64
-        - name: name
-          dtype: string
-        - name: supercategory
-          dtype: string
-    - name: image # This feature is added by the Hub for image datasets
-      dtype: image
-  configs:
-    - config_name: coco_far_voc
-    - config_name: coco_farther_bdd
-    - config_name: coco_near_voc
-    - config_name: oi_far_voc
-    - config_name: oi_farther_bdd
-    - config_name: oi_near_voc
----
-
-# FMIYC (Find Me If You Can) Dataset
-
-This dataset is designed for Out-Of-Distribution (OOD) Object Detection tasks. It comprises images sourced from COCO and OpenImages, with annotations provided in the COCO JSON format. The dataset is structured into several configurations based on the image source and perspective.
-
-## Dataset Structure
-
-### Data Instances
-
-Each instance corresponds to an image and its associated object annotations.
-The features include:
-*   `file_name`: (string) The original filename of the image.
-*   `image_id`: (string) Unique identifier for the image.
-*   `height`: (float32) Height of the image in pixels.
-*   `width`: (float32) Width of the image in pixels.
-*   `dataset_origin`: (string) Source dataset ("COCO" or "OpenImages").
-*   `distance_category`: (string) Perspective ("near", "far", "farther").
-*   `objects`: A list of dictionaries, where each dictionary represents an annotated object:
-    *   `id`: (string) Unique annotation ID.
-    *   `area`: (float32) Area of the bounding box.
-    *   `bbox_x`: (float32) The x-coordinate of the top-left corner of the bounding box.
-    *   `bbox_y`: (float32) The y-coordinate of the top-left corner of the bounding box.
-    *   `bbox_width`: (float32) The width of the bounding box.
-    *   `bbox_height`: (float32) The height of the bounding box.
-    *   `category_id`: (int64) ID of the object category.
-*   `categories`: A list of dictionaries defining all categories for the current configuration:
-    *   `id`: (int64) Unique category ID.
-    *   `name`: (string) Category name.
-    *   `supercategory`: (string) Supercategory name.
-*   `image`: The image object itself (added by the Hugging Face datasets library).
-
-
-### Dataset Configurations
-*   `coco_far_voc`
-*   `coco_farther_bdd`
-*   `coco_near_voc`
-*   `oi_far_voc`
-*   `oi_farther_bdd`
-*   `oi_near_voc`
-
-Each configuration is processed into a `train` split.
-
-## Citation
-```bibtex
-@misc{Montoya_FindMeIfYouCan_YYYY,
-  author    = {Montoya, Daniel and Bouguerra, Aymen and Gomez-Villa, Alexandra and Arnez, Fabio},
-  title     = {FindMeIfYouCan: Bringing Open Set metrics to near, far and farther Out-of-Distribution Object detection},
-  year      = {YYYY}, # Replace YYYY
-}
-**Steps to Try This:**
-
-1.  **Replace your `restructure_fmiyc.py`** with the updated version above.
-2.  **Run it locally:** `python restructure_fmiyc.py`. This will create a new `FMIYC_restructured` directory with `metadata.jsonl` files containing the flattened bbox structure.
-3.  **Replace your `README.md`** in your local Git repository with the updated version above.
-4.  **Upload to Hugging Face Hub:**
-    *   You'll need to `git add README.md`.
-    *   You'll also need to re-add the `FMIYC_restructured` directory if you had removed it from Git tracking, or if its contents changed significantly. If it was already tracked, Git will pick up the changes in the `metadata.jsonl` files.
-    *   A clean way if you're unsure about Git state for `FMIYC_restructured`:
-        ```bash
-        # Assuming FMIYC_restructured is in your .gitignore or wasn't tracked properly
-        # If it WAS tracked, these 'rm' commands are for the files on disk, not git history yet
-        # Make sure you have a backup or can regenerate FMIYC_restructured if needed!
-
-        # Remove the old restructured data from your Git tracking (if it was added)
-        # git rm -r --cached FMIYC_restructured # Only removes from index, keeps files
-        # git commit -m "Prepare for new restructured data"
-
-        # Now, ensure FMIYC_restructured (newly generated) is ready to be added
-        git add FMIYC_restructured/
-        git add README.md
-        git commit -m "Update data with flattened bbox and revised README"
-        git push origin main # Or your branch
-        ```
-    *   **If `FMIYC_restructured` was never properly tracked with LFS for its images:** You'd need to ensure LFS is tracking image files *within* `FMIYC_restructured/*/*.jpg` etc., then `git add` and commit.
-        The simplest for a clean upload of `FMIYC_restructured`:
-        1.  Delete the old `FMIYC_restructured` from your Hub repo (via web UI or `git rm` and push).
-        2.  Run the new `restructure_fmiyc.py` locally.
-        3.  In your local repo:
-            ```bash
-            # Make sure LFS is tracking images inside the new FMIYC_restructured
-            # (if not already covered by a broader rule)
-            # git lfs track "FMIYC_restructured/**/*.jpg" 
-            # git lfs track "FMIYC_restructured/**/*.png"
-            # git add .gitattributes # if changed
-
-            git add README.md
-            git add FMIYC_restructured/
-            git commit -m "Upload restructured data with flattened bbox and new README"
-            git push origin main
-            ```
-
-This approach (flattening `bbox`) directly addresses the most likely point of failure in the Croissant generator by removing the complex nested sequence that it seems to be mishandling. This has the highest chance of making the "Records Generation Test" pass if you are sticking to the `metadata.jsonl` + `README.md` upload method.