Hugging Face's logo

thenukegun10x
/
PLantDetect-WA

Image Classification
LiteRT
LiteRT
ONNX
English
vision
botany
western-australia
dinov3
mixture-of-experts
adaround
fp8
int8
android
biodiversity
flora
Model card Files
xet
 Community
thenukegun10x commited on
Commit
9031464
·
verified ·
1 Parent(s): 0aa8ad7

Add direct download links, accurate AdaRound FP8 vs INT8 quantization specs, and clean filenames

Browse files
Files changed (1) hide show
  1. README.md +25 -18
README.md CHANGED
@@ -50,35 +50,42 @@ In the field, plant species can be difficult to distinguish from a single photo.
50
 
51
  ---
52
 
53
- ## 2. Complete Model Artifact Catalog
54
 
55
  ### A. Mobile Edge & Embedded Models (Dense ViT-Base)
56
  *Optimized for Samsung Galaxy S24 (Exynos 2400 / Snapdragon 8 Gen 3), iOS, Windows DirectML, and Raspberry Pi.*
57
 
58
- | Model File | Model Type | Runtime Target | Format / Precision | File Size | Top-1 (1-View) | Top-1 (4-View) | Target Hardware |
59
  | :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- |
60
- | **`PlantDetect-Dense-INT8-AdaRound.safetensors`** | Single-Shot | PyTorch / Python | **AdaRound INT8 (W8A16)** | **`93.3 MB`** | 83.06% | 97.35% | Ultra-compact Python edge |
61
- | **`PlantDetect-Dense-FP8-AdaRound.safetensors`** | Single-Shot | PyTorch / Python | **Mixed AdaRound FP8** | **`92.7 MB`** | 83.06% | 97.35% | GPU / Python lightweight |
62
- | **`PlantDetect-Dense-BF16.safetensors`** | Single-Shot | PyTorch / Python | **Full `bfloat16`** | **`181.5 MB`** | 83.06% | 97.35% | Dense PyTorch reference |
63
- | **`PlantDetect-Dense-4View-INT8.safetensors`** | **4-View Distilled** | PyTorch / Python | **AdaRound INT8 (W8A16)** | **`93.3 MB`** | 85.10% | **99.10%** | Dedicated multi-photo mobile edge |
64
- | **`PlantDetect-Dense-4View-BF16.safetensors`** | **4-View Distilled** | PyTorch / Python | **Full `bfloat16`** | **`181.5 MB`** | 85.10% | **99.10%** | Lossless 4-view dense reference |
65
- | **`onnx/PlantDetect-Dense-INT8-AdaRound.onnx`** | Single-Shot | ONNX Runtime | **Mixed-Precision INT8** | **`106.3 MB`** | 83.06% | 97.35% | Android, iOS CoreML, Windows |
66
- | **`onnx/PlantDetect-Dense-FP32.onnx`** | Single-Shot | ONNX Runtime | **Full FP32** | `363.3 MB` | 83.06% | 97.35% | Standard reference ONNX |
67
- | **`litert/PlantDetect-Dense-INT8.tflite`** | Single-Shot | Google LiteRT (TFLite) | **Full Integer INT8** | **`387.0 MB`** | 83.06% | 97.35% | Samsung S24 NPU acceleration |
68
- | **`litert/PlantDetect-Dense-FP16.tflite`** | Single-Shot | Google LiteRT (TFLite) | **Float16** | **`196.2 MB`** | 83.06% | 97.35% | Mobile GPU / XNNPACK CPU |
 
 
 
 
 
69
 
70
  ---
71
 
72
  ### B. Server / Desktop GPU Models (Mixture-of-Experts)
73
 
74
- | Model File | Model Type | Architecture | Format | Size | Top-1 (1-View) | Top-1 (4-View) | Primary Use Case |
75
  | :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- |
76
- | **`PlantDetect-FP8-AdaRound.safetensors`** | Single-Shot | MoE++ (16 FFNs) | **Mixed FP8 + BF16** | **`104.3 MB`** | **89.21%** | **96.40%** | **Fastest, ultra-compact GPU server inference** |
77
- | **`PlantDetect-BF16.safetensors`** | Single-Shot | MoE++ (16 FFNs) | **Full `bfloat16`** | **`205.2 MB`** | **89.21%** | **96.40%** | Lossless baseline reference |
78
- | **`PlantDetect-4View-FP8-AdaRound.safetensors`** | **4-View Distilled** | MoE++ 4-View | **Mixed FP8 + BF16** | **`104.3 MB`** | 85.06% | **99.20%** | Dedicated multi-photo teacher |
79
- | **`PlantDetect-4View-BF16.safetensors`** | **4-View Distilled** | MoE++ 4-View | **Full `bfloat16`** | **`205.2 MB`** | 85.06% | **99.20%** | Lossless multi-photo baseline |
80
-
81
- > **Note on 4-View Weights:** The `PlantDetect-4View` models have dedicated fine-tuned weights trained via **Grouped Multi-View Knowledge Distillation** across 39,502 observation groups ($K=4$). They explicitly learn cross-angle feature synergy between flowers, foliage, habit, and fruit.
 
 
82
 
83
  ---
84
 
 
50
 
51
  ---
52
 
53
+ ## 2. Complete Model Artifact Catalog (Direct Download Links)
54
 
55
  ### A. Mobile Edge & Embedded Models (Dense ViT-Base)
56
  *Optimized for Samsung Galaxy S24 (Exynos 2400 / Snapdragon 8 Gen 3), iOS, Windows DirectML, and Raspberry Pi.*
57
 
58
+ | Model File | Model Type | Runtime Target | Quantization Method | File Size | Top-1 (1-View) | Top-1 (4-View) | Target Hardware |
59
  | :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- |
60
+ | [**`PlantDetect-Dense-INT8.safetensors`**](https://huggingface.co/thenukegun10x/wa-plant-identifier/resolve/main/PlantDetect-Dense-INT8.safetensors) | Single-Shot | PyTorch / Python | **Standard W8A16 INT8** | **`93.3 MB`** | 83.06% | 97.35% | Ultra-compact Python edge |
61
+ | [**`PlantDetect-Dense-FP8.safetensors`**](https://huggingface.co/thenukegun10x/wa-plant-identifier/resolve/main/PlantDetect-Dense-FP8.safetensors) | Single-Shot | PyTorch / Python | **AdaRound FP8 (E4M3)** | **`92.7 MB`** | 83.06% | 97.35% | GPU / Python lightweight |
62
+ | [**`PlantDetect-Dense-BF16.safetensors`**](https://huggingface.co/thenukegun10x/wa-plant-identifier/resolve/main/PlantDetect-Dense-BF16.safetensors) | Single-Shot | PyTorch / Python | **Full `bfloat16`** | **`181.5 MB`** | 83.06% | 97.35% | Dense PyTorch reference |
63
+ | [**`PlantDetect-Dense-4View-INT8.safetensors`**](https://huggingface.co/thenukegun10x/wa-plant-identifier/resolve/main/PlantDetect-Dense-4View-INT8.safetensors) | **4-View Distilled** | PyTorch / Python | **Standard W8A16 INT8** | **`93.3 MB`** | 82.13% | **97.60%** | Dedicated multi-photo mobile edge |
64
+ | [**`PlantDetect-Dense-4View-FP8.safetensors`**](https://huggingface.co/thenukegun10x/wa-plant-identifier/resolve/main/PlantDetect-Dense-4View-FP8.safetensors) | **4-View Distilled** | PyTorch / Python | **AdaRound FP8 (E4M3)** | **`92.7 MB`** | 82.13% | **97.60%** | Dedicated multi-photo GPU edge |
65
+ | [**`PlantDetect-Dense-4View-BF16.safetensors`**](https://huggingface.co/thenukegun10x/wa-plant-identifier/resolve/main/PlantDetect-Dense-4View-BF16.safetensors) | **4-View Distilled** | PyTorch / Python | **Full `bfloat16`** | **`181.5 MB`** | 82.13% | **97.60%** | Lossless 4-view dense reference |
66
+ | [**`onnx/PlantDetect-Dense-INT8.onnx`**](https://huggingface.co/thenukegun10x/wa-plant-identifier/resolve/main/onnx/PlantDetect-Dense-INT8.onnx) | Single-Shot | ONNX Runtime | **Mixed-Precision INT8** | **`106.3 MB`** | 83.06% | 97.35% | Android, iOS CoreML, Windows |
67
+ | [**`onnx/PlantDetect-Dense-FP32.onnx`**](https://huggingface.co/thenukegun10x/wa-plant-identifier/resolve/main/onnx/PlantDetect-Dense-FP32.onnx) | Single-Shot | ONNX Runtime | **Full FP32** | `363.3 MB` | 83.06% | 97.35% | Standard reference ONNX |
68
+ | [**`onnx/PlantDetect-Dense-4View-INT8.onnx`**](https://huggingface.co/thenukegun10x/wa-plant-identifier/resolve/main/onnx/PlantDetect-Dense-4View-INT8.onnx) | **4-View Distilled** | ONNX Runtime | **Mixed-Precision INT8** | **`106.3 MB`** | 82.13% | **97.60%** | Multi-photo ONNX Runtime |
69
+ | [**`onnx/PlantDetect-Dense-4View-FP32.onnx`**](https://huggingface.co/thenukegun10x/wa-plant-identifier/resolve/main/onnx/PlantDetect-Dense-4View-FP32.onnx) | **4-View Distilled** | ONNX Runtime | **Full FP32** | `363.3 MB` | 82.13% | **97.60%** | Multi-photo reference ONNX |
70
+ | [**`litert/PlantDetect-Dense-INT8.tflite`**](https://huggingface.co/thenukegun10x/wa-plant-identifier/resolve/main/litert/PlantDetect-Dense-INT8.tflite) | Single-Shot | Google LiteRT (TFLite) | **Full Integer INT8** | **`387.0 MB`** | 83.06% | 97.35% | Samsung S24 NPU acceleration |
71
+ | [**`litert/PlantDetect-Dense-FP16.tflite`**](https://huggingface.co/thenukegun10x/wa-plant-identifier/resolve/main/litert/PlantDetect-Dense-FP16.tflite) | Single-Shot | Google LiteRT (TFLite) | **Float16** | **`196.2 MB`** | 83.06% | 97.35% | Mobile GPU / XNNPACK CPU |
72
+ | [**`litert/PlantDetect-Dense-4View-INT8.tflite`**](https://huggingface.co/thenukegun10x/wa-plant-identifier/resolve/main/litert/PlantDetect-Dense-4View-INT8.tflite) | **4-View Distilled** | Google LiteRT (TFLite) | **Full Integer INT8** | **`385.8 MB`** | 82.13% | **97.60%** | Samsung S24 Multi-View NPU |
73
+ | [**`litert/PlantDetect-Dense-4View-FP16.tflite`**](https://huggingface.co/thenukegun10x/wa-plant-identifier/resolve/main/litert/PlantDetect-Dense-4View-FP16.tflite) | **4-View Distilled** | Google LiteRT (TFLite) | **Float16** | **`195.8 MB`** | 82.13% | **97.60%** | Multi-View Mobile GPU |
74
 
75
  ---
76
 
77
  ### B. Server / Desktop GPU Models (Mixture-of-Experts)
78
 
79
+ | Model File | Model Type | Architecture | Quantization Method | Size | Top-1 (1-View) | Top-1 (4-View) | Primary Use Case |
80
  | :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- |
81
+ | [**`PlantDetect-FP8.safetensors`**](https://huggingface.co/thenukegun10x/wa-plant-identifier/resolve/main/PlantDetect-FP8.safetensors) | Single-Shot | MoE++ (16 FFNs) | **AdaRound FP8 (E4M3)** | **`104.3 MB`** | **89.21%** | **96.40%** | **Fastest, ultra-compact GPU server inference** |
82
+ | [**`PlantDetect-BF16.safetensors`**](https://huggingface.co/thenukegun10x/wa-plant-identifier/resolve/main/PlantDetect-BF16.safetensors) | Single-Shot | MoE++ (16 FFNs) | **Full `bfloat16`** | **`205.2 MB`** | **89.21%** | **96.40%** | Lossless baseline reference |
83
+ | [**`PlantDetect-4View-FP8.safetensors`**](https://huggingface.co/thenukegun10x/wa-plant-identifier/resolve/main/PlantDetect-4View-FP8.safetensors) | **4-View Distilled** | MoE++ 4-View | **AdaRound FP8 (E4M3)** | **`104.3 MB`** | 85.06% | **99.20%** | Dedicated multi-photo teacher (FP8) |
84
+ | [**`PlantDetect-4View-BF16.safetensors`**](https://huggingface.co/thenukegun10x/wa-plant-identifier/resolve/main/PlantDetect-4View-BF16.safetensors) | **4-View Distilled** | MoE++ 4-View | **Full `bfloat16`** | **`205.2 MB`** | 85.06% | **99.20%** | Lossless multi-photo baseline |
85
+
86
+ > **Quantization Architecture Details:**
87
+ > * **FP8 Models:** Quantized using **AdaRound FP8** with quadratic activation reconstruction loss $\min \|Wx - \tilde{W}(V)x\|_2^2$ over non-uniform E4M3 discrete grids (100% Top-1 FP32 match).
88
+ > * **INT8 Models:** Quantized using **Symmetric Per-Channel W8A16 Dynamic Range Quantization**, delivering **`43.81 dB` SQNR** and **`0.999985` cosine similarity**.
89
 
90
  ---
91