Upload preprocessor tests

test_preprocessor_config.py

@@ -0,0 +1,125 @@
+import marimo
+
+__generated_with = "0.17.0"
+app = marimo.App()
+
+
+@app.cell
+def _():
+    from open_clip import create_model_from_pretrained
+    import torch
+    from transformers import CLIPImageProcessor, AutoImageProcessor
+    from urllib.request import urlopen
+    from PIL import Image
+    return (
+        AutoImageProcessor,
+        CLIPImageProcessor,
+        Image,
+        create_model_from_pretrained,
+        torch,
+        urlopen,
+    )
+
+
+@app.cell
+def _(Image, urlopen):
+    image = Image.open(urlopen(
+        # 'https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'
+        'https://media.gettyimages.com/id/1309385864/photo/las-vegas-strip-skyline-landscaped-ultra-wide-shot-at-night.jpg?s=1024x1024&w=gi&k=20&c=bBxkjq8FoK2bPKYLtqbJMCgpsvjUm-vI7-yw04cq7AU='
+    ))
+    image.size
+    return (image,)
+
+
+@app.cell
+def _():
+    model = "apple/DFN5B-CLIP-ViT-H-14"
+    return (model,)
+
+
+@app.cell
+def _(
+    AutoImageProcessor,
+    CLIPImageProcessor,
+    create_model_from_pretrained,
+    model,
+):
+    try:
+        hf_pre = AutoImageProcessor.from_pretrained(model)
+    except:
+        print("Auto image processor not found")
+        hf_pre = CLIPImageProcessor()
+    hf_pre_fix = CLIPImageProcessor(
+        do_center_crop=False,
+        do_normalize=True,
+        do_resize=True,
+        feature_extractor_type="CLIPFeatureExtractor",
+        image_mean=[
+            0.48145466,
+            0.4578275,
+            0.40821073
+        ],
+        image_std=[
+            0.26862954,
+            0.26130258,
+            0.27577711
+        ],
+        size={"width": 224, "height": 224},
+        do_convert_rgb=True
+    )
+    _, preprocess = create_model_from_pretrained(f'hf-hub:{model}')
+    preprocess
+    return hf_pre, hf_pre_fix, preprocess
+
+
+@app.cell
+def _(hf_pre, hf_pre_fix, image, preprocess, torch):
+    hf_res_fix = torch.tensor(hf_pre_fix(images=image)["pixel_values"]).squeeze()
+    hf_res = torch.tensor(hf_pre(images=image)["pixel_values"]).squeeze()
+    op_res = preprocess(image)
+    op_res.shape, hf_res.shape
+    return hf_res, hf_res_fix, op_res
+
+
+@app.cell
+def _(hf_res, op_res):
+    (hf_res == op_res).all()
+    return
+
+
+@app.cell
+def _(hf_res_fix, op_res):
+    (hf_res_fix == op_res).all()
+    return
+
+
+@app.cell
+def _(hf_res, hf_res_fix, image, op_res):
+    # view the images
+    import matplotlib.pyplot as plt
+    fig, axs = plt.subplots(1, 4, figsize=(15, 5))
+    axs[0].imshow(image)
+    axs[0].set_title("Original Image")
+    axs[1].imshow(op_res.permute(1, 2, 0).numpy())
+    axs[1].set_title("OpenCLIP Preprocessed Image")
+    axs[2].imshow(hf_res.permute(1, 2, 0).numpy())
+    axs[2].set_title("HuggingFace Preprocessed Image")
+    axs[3].imshow(hf_res_fix.permute(1, 2, 0).numpy())
+    axs[3].set_title("HuggingFace FIXED Preprocessed Image")
+    plt.show()
+    return
+
+
+@app.cell
+def _(hf_pre_fix):
+    hf_pre_fix.to_json_file("test.json")
+    return
+
+
+@app.cell
+def _():
+    return
+
+
+if __name__ == "__main__":
+    app.run()