Instructions to use Datadragon110/freshwater-quantization-models with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Keras
How to use Datadragon110/freshwater-quantization-models with Keras:
# !pip install -U keras tensorflow huggingface_hub # Keras needs TensorFlow installed to read "hf://" paths, so the tensorflow backend is selected here; # "jax" and "torch" also work for computation once TensorFlow is installed. import os os.environ["KERAS_BACKEND"] = "tensorflow" import keras model = keras.saving.load_model("hf://Datadragon110/freshwater-quantization-models") - Notebooks
- Google Colab
- Kaggle
Download paper_claim_check.csv from Datadragon110/freshwater-quantization-models: direct link, hf CLI and curl.
- Browser
- Download file 1.82 kB
-
https://huggingface.co/Datadragon110/freshwater-quantization-models/resolve/main/paper_claim_check.csv
- Command line
-
hf download hf://Datadragon110/freshwater-quantization-models/paper_claim_check.csv
-
curl -L -o paper_claim_check.csv https://huggingface.co/Datadragon110/freshwater-quantization-models/resolve/main/paper_claim_check.csv
1.82 kB
| claim,paper,measured,delta,tol,status | |
| Train-folder images (paper: 1485 + 262),1747.0,1750.0,3.0,0.0,MISMATCH | |
| Test-folder images,697.0,700.0,3.0,0.0,MISMATCH | |
| Test files that also exist in Train (should be 0),0.0,700.0,700.0,0.0,MISMATCH | |
| "Test images with a near-duplicate in Train, % (should be ~0)",0.0,100.0,100.0,1.0,MISMATCH | |
| classes,7.0,7.0,0.0,0.0,MATCH | |
| EfficientNetB0 total params,4058538.0,4058538.0,0.0,0.0,MATCH | |
| EfficientNetB0 frozen params,4049571.0,4049571.0,0.0,0.0,MATCH | |
| MobileNetV3Small total params,445039.0,445039.0,0.0,0.0,MATCH | |
| MobileNetV3Small frozen params,441000.0,441000.0,0.0,0.0,MATCH | |
| "EfficientNetB0 fp32 acc, PAPER protocol (leaky Test/)",90.24,93.286,3.046,3.0,MISMATCH | |
| "EfficientNetB0 INT8 acc, PAPER protocol (leaky Test/)",69.44,79.143,9.703,3.0,MISMATCH | |
| "EfficientNetB0 fp32 acc, LEAKAGE-FREE grouped CV",90.24,88.004,-2.236,1.993,MISMATCH | |
| "EfficientNetB0 INT8 acc, LEAKAGE-FREE grouped CV",69.44,74.176,4.736,6.209,MATCH | |
| "MobileNetV3Small fp32 acc, PAPER protocol (leaky Test/)",81.78,83.714,1.934,3.0,MATCH | |
| "MobileNetV3Small INT8 acc, PAPER protocol (leaky Test/)",59.08,65.714,6.634,3.0,MISMATCH | |
| "MobileNetV3Small fp32 acc, LEAKAGE-FREE grouped CV",81.78,79.287,-2.493,4.511,MATCH | |
| "MobileNetV3Small INT8 acc, LEAKAGE-FREE grouped CV",59.08,60.007,0.927,10.445,MATCH | |
| EfficientNetB0 fp32 size (KiB),15700.1,15700.116,0.016,157.001,MATCH | |
| EfficientNetB0 INT8 size (KiB),4808.1,4808.125,0.025,48.081,MATCH | |
| MobileNetV3Small fp32 size (KiB),1690.8,1690.846,0.046,16.908,MATCH | |
| MobileNetV3Small INT8 size (KiB),603.7,603.717,0.017,6.037,MATCH | |
| size reduction EffNet fp32 -> INT8 (x),3.3,3.265,-0.035,0.1,MATCH | |
| size reduction MNv3 fp32 -> INT8 (x),2.8,2.801,0.001,0.1,MATCH | |
| size reduction EffNet fp32 -> MNv3 INT8 (x),26.0,26.006,0.006,0.5,MATCH | |
| param ratio EffNet / MNv3 (x),9.1,9.12,0.02,0.1,MATCH | |