kjcpc
/

ML-Sharp-QNN

@@ -24,13 +24,13 @@ This is a **Qualcomm QNN (Qualcomm Neural Network)** conversion of [apple/Sharp]
 **Converted models (5 DLC files):**
-| File | Model | Description |
-|------|-------|-------------|
-| `pe.dlc` | Patch Encoder | Encodes 35 image patches (384×384) |
-| `ie.dlc` | Image Encoder | Encodes the full image (1536×1536) |
-| `rest_a.dlc` | REST Segment A | Feature Fusion (6 inputs → 6 outputs) |
-| `rest_b.dlc` | REST Segment B | Disparity Estimation |
-| `rest_c.dlc` | REST Segment C | Gaussian Delta Prediction |
 ---
@@ -45,7 +45,7 @@ This software project accompanies the research paper: _Sharp Monocular View Synt
 by _Lars Mescheder, Wei Dong, Shiwei Li, Xuyang Bai, Marcel Santos, Peiyun Hu, Bruno Lecouat, Mingmin Zhen, Amaël Delaunoy,
 Tian Fang, Yanghai Tsin, Stephan Richter and Vladlen Koltun_.
-![Teaser](https://huggingface.co/kjcpc/ML-Sharp-QNN/resolve/main/data/teaser.jpg)
 We present SHARP, an approach to photorealistic view synthesis from a single image. Given a single photograph, SHARP regresses the parameters of a 3D Gaussian representation of the depicted scene. This is done in less than a second on a standard GPU via a single feedforward pass through a neural network. The 3D Gaussian representation produced by SHARP can then be rendered in real time, yielding high-resolution photorealistic images for nearby views. The representation is metric, with absolute scale, supporting metric camera movements. Experimental results demonstrate that SHARP delivers robust zero-shot generalization across datasets. It sets a new state of the art on multiple datasets, reducing LPIPS by 25–34% and DISTS by 21–43% versus the best prior model, while lowering the synthesis time by three orders of magnitude.
@@ -152,7 +152,7 @@ Input Image
 ## Usage with the Android App
-1. Download all 5 `.dlc` files from this repository
 2. Place them in the app's model directory under `dlc/`:
    ```
    <modelRoot>/dlc/pe.dlc

 **Converted models (5 DLC files):**
+| File | Model | Size | Description |
+|------|-------|------|-------------|
+| `dlc/w8a16/pe.dlc` | Patch Encoder | 306 MB | Encodes 35 image patches (384×384) |
+| `dlc/w8a16/ie.dlc` | Image Encoder | 306 MB | Encodes the full image (1536×1536) |
+| `dlc/w8a16/rest_a.dlc` | REST Segment A | 33 MB | Feature Fusion (6 inputs → 6 outputs) |
+| `dlc/w8a16/rest_b.dlc` | REST Segment B | 6 MB | Disparity Estimation |
+| `dlc/w8a16/rest_c.dlc` | REST Segment C | 8 MB | Gaussian Delta Prediction |
 ---
 by _Lars Mescheder, Wei Dong, Shiwei Li, Xuyang Bai, Marcel Santos, Peiyun Hu, Bruno Lecouat, Mingmin Zhen, Amaël Delaunoy,
 Tian Fang, Yanghai Tsin, Stephan Richter and Vladlen Koltun_.
+![](https://huggingface.co/kjcpc/ML-Sharp-QNN/resolve/main/data/teaser.jpg)
 We present SHARP, an approach to photorealistic view synthesis from a single image. Given a single photograph, SHARP regresses the parameters of a 3D Gaussian representation of the depicted scene. This is done in less than a second on a standard GPU via a single feedforward pass through a neural network. The 3D Gaussian representation produced by SHARP can then be rendered in real time, yielding high-resolution photorealistic images for nearby views. The representation is metric, with absolute scale, supporting metric camera movements. Experimental results demonstrate that SHARP delivers robust zero-shot generalization across datasets. It sets a new state of the art on multiple datasets, reducing LPIPS by 25–34% and DISTS by 21–43% versus the best prior model, while lowering the synthesis time by three orders of magnitude.
 ## Usage with the Android App
+1. Download all 5 `.dlc` files from `dlc/w8a16/` of this repository
 2. Place them in the app's model directory under `dlc/`:
    ```
    <modelRoot>/dlc/pe.dlc