Update README.md

README.md

@@ -1,7 +1,25 @@
+---
+license: mit
+library_name: pytorch
+tags:
+- computer-vision
+- image-segmentation
+- edge-detection
+- line-art
+- anime
+datasets:
+- custom
+metrics:
+- dice
+- iou
+pipeline_tag: image-segmentation
+---
+
 # Anime Line Art Extraction Segmentation Model
 <div style="display:flex;">
 <img src="https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/6kCBB668giXjJoCLXAzfy.png" width="50%">
-<img src="https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/Q5pwvCCAfhl5ctqgsVEPa.png" width="50%"></div>
+<img src="https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/Q5pwvCCAfhl5ctqgsVEPa.png" width="50%">
+</div>
 
 ## Model Description
 
@@ -35,7 +53,7 @@ Frameworks used:
 
 Since no pretrained model exists specifically for anime line extraction, the model was trained using a custom dataset and automatically generated edge masks.
 
-![6f1b30d9-9aed-4577-bedc-b2c367ccda9a](https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/h06ej-ODkw5tDAx3X6KfL.png)
+<img src="https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/h06ej-ODkw5tDAx3X6KfL.png" width="80%">
 
 ### Intended Use Cases
 Potential applications include:
@@ -198,7 +216,7 @@ IoU: ~0.21
 
 These metrics indicate that the model is able to detect meaningful edge structures but struggles with extremely thin line details.
 
-![a17d7a6a-7832-46b2-a37a-3d1a09f78f2d](https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/zvCczs-TB241YW4FuVIOF.png)
+<img src="https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/zvCczs-TB241YW4FuVIOF.png" width="35%">
 
 ## Key Observations
 
@@ -208,7 +226,7 @@ What worked well:
 - Captured hair boundaries
 - Detected facial structures
 
-![4b2f011b-1176-442b-b210-ebfe9d46295f](https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/dlkCHCrPtBJPvy7sGSc8j.png)
+<img src="https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/dlkCHCrPtBJPvy7sGSc8j.png" width="75%">
 
 Failure cases:
 
@@ -216,10 +234,10 @@ Failure cases:
 - Dark scenes
 - Shading lines interpreted as edges
 - Excessive background detail
-
-![177d655b-cbe4-4cfa-818e-73f3067f2bc9](https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/Hi0LQhIQZvWlAd_44o88H.png)
-![780eba9f-d4da-4b13-8ad6-88d65b1e64bb](https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/KnIMiKkePB9aDNausGNDp.png)
-
+<div>
+<img src="https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/Hi0LQhIQZvWlAd_44o88H.png" width="75%">
+<img src="https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/KnIMiKkePB9aDNausGNDp.png" width="75%">
+</div>
 These results show that the model learned meaningful edge structures despite the noisy annotations generated from Canny edge detection.
 
 ## Visual Examples
@@ -243,11 +261,11 @@ However, the task presents several challenges:
 3. Low contrast scenes reduce edge detectability
 
 Because the model was only trained for 30 epochs, additional training may improve performance. However, improving annotation quality or training at higher resolution would likely have a larger impact.
-
-![4e5dafe3-942a-4cb1-af0f-de273cdb638a](https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/ru-gguNfSDzxbCXT6kmeS.png)
-![05e59579-9f5f-43e9-85c4-48796256c7a3](https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/wT3J4LSINPNHVVjLUaqcR.png)
-![b7ac9794-9a14-4c83-a3d4-44dffd961b61](https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/PL9-L1MHMEhqmNxY4WQkm.png)
-
+<div style="display:flex;">
+<img src="https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/PL9-L1MHMEhqmNxY4WQkm.png" width="50%">
+<img src="https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/ru-gguNfSDzxbCXT6kmeS.png" width="50%">
+</div>
+<img src="https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/wT3J4LSINPNHVVjLUaqcR.png" width="50%">
 --------------------------------------------------
 
 # Limitations and Biases
@@ -319,12 +337,12 @@ Additional research directions include:
 
 - Object detection models for automatic removal of occlusions
 
-![79f0da0a-4e82-4145-a13f-f4b4cef2ea7d](https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/NKCNnMBSAzzhAPjaZiX9y.png)
+<img src="https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/NKCNnMBSAzzhAPjaZiX9y.png" width="50%">
 
 - Line art upscaling techniques
 
-![f95b019a-cd46-495b-841e-4d860b9ea6d9](https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/3o1v3xkefzt04DJi2wobi.png)
+<img src="https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/9cisCYIkU_y45UJtJRNcE.png" width="50%">
 
 - Using detected edges for stitching animation panning shots
 
-![unnamed](https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/ZDIrGENzx4oy-Vj_jyQMa.gif)
+<img src="https://cdn-uploads.huggingface.co/production/uploads/6972a2622ef5ed3b50628995/ZDIrGENzx4oy-Vj_jyQMa.gif" width="50%">