"second commit"

app.py

@@ -1,18 +1,14 @@
 import gradio as gr
-
 from matplotlib import gridspec
 import matplotlib.pyplot as plt
 import numpy as np
 from PIL import Image
-import tensorflow as tf
-from transformers import SegformerFeatureExtractor, TFSegformerForSemanticSegmentation
+import torch
+from transformers import AutoImageProcessor, AutoModelForSemanticSegmentation
 
-feature_extractor = SegformerFeatureExtractor.from_pretrained(
-    "nvidia/segformer-b5-finetuned-ade-640-640"
-)
-model = TFSegformerForSemanticSegmentation.from_pretrained(
-    "nvidia/segformer-b5-finetuned-ade-640-640"
-)
+MODEL_ID = "nvidia/segformer-b5-finetuned-ade-640-640"
+processor = AutoImageProcessor.from_pretrained(MODEL_ID)
+model = AutoModelForSemanticSegmentation.from_pretrained(MODEL_ID)
 
 def ade_palette():
     """ADE20K palette that maps each class to RGB values."""
@@ -38,34 +34,32 @@ def ade_palette():
     ]
 
 labels_list = []
-
-with open(r'labels.txt', 'r') as fp:
+with open("labels.txt", "r", encoding="utf-8") as fp:
     for line in fp:
-        labels_list.append(line[:-1])
+        labels_list.append(line.rstrip("\n"))
 
-colormap = np.asarray(ade_palette())
+colormap = np.asarray(ade_palette(), dtype=np.uint8)
 
 def label_to_color_image(label):
     if label.ndim != 2:
         raise ValueError("Expect 2-D input label")
-
     if np.max(label) >= len(colormap):
         raise ValueError("label value too large.")
     return colormap[label]
 
-def draw_plot(pred_img, seg):
+def draw_plot(pred_img, seg_np):
     fig = plt.figure(figsize=(20, 15))
-
     grid_spec = gridspec.GridSpec(1, 2, width_ratios=[6, 1])
 
     plt.subplot(grid_spec[0])
     plt.imshow(pred_img)
     plt.axis('off')
+
     LABEL_NAMES = np.asarray(labels_list)
     FULL_LABEL_MAP = np.arange(len(LABEL_NAMES)).reshape(len(LABEL_NAMES), 1)
     FULL_COLOR_MAP = label_to_color_image(FULL_LABEL_MAP)
 
-    unique_labels = np.unique(seg.numpy().astype("uint8"))
+    unique_labels = np.unique(seg_np.astype("uint8"))
     ax = plt.subplot(grid_spec[1])
     plt.imshow(FULL_COLOR_MAP[unique_labels].astype(np.uint8), interpolation="nearest")
     ax.yaxis.tick_right()
@@ -74,37 +68,43 @@ def draw_plot(pred_img, seg):
     ax.tick_params(width=0.0, labelsize=25)
     return fig
 
-def sepia(input_img):
-    input_img = Image.fromarray(input_img)
-
-    inputs = feature_extractor(images=input_img, return_tensors="tf")
-    outputs = model(**inputs)
-    logits = outputs.logits
+def run_inference(input_img):
+    # input: numpy array from gradio -> PIL
+    img = Image.fromarray(input_img.astype(np.uint8)) if isinstance(input_img, np.ndarray) else input_img
+    if img.mode != "RGB":
+        img = img.convert("RGB")
 
-    logits = tf.transpose(logits, [0, 2, 3, 1])
-    logits = tf.image.resize(
-        logits, input_img.size[::-1]
-    )  # We reverse the shape of `image` because `image.size` returns width and height.
-    seg = tf.math.argmax(logits, axis=-1)[0]
+    inputs = processor(images=img, return_tensors="pt")
+    with torch.no_grad():
+        outputs = model(**inputs)
+        logits = outputs.logits  # (1, C, h/4, w/4)
 
-    color_seg = np.zeros(
-        (seg.shape[0], seg.shape[1], 3), dtype=np.uint8
-    )  # height, width, 3
-    for label, color in enumerate(colormap):
-        color_seg[seg.numpy() == label, :] = color
+    # resize to original
+    upsampled = torch.nn.functional.interpolate(
+        logits, size=img.size[::-1], mode="bilinear", align_corners=False
+    )
+    seg = upsampled.argmax(dim=1)[0].cpu().numpy().astype(np.uint8)  # (H,W)
 
-    # Show image + mask
-    pred_img = np.array(input_img) * 0.5 + color_seg * 0.5
-    pred_img = pred_img.astype(np.uint8)
+    # colorize & overlay
+    color_seg = colormap[seg]                                # (H,W,3)
+    pred_img = (np.array(img) * 0.5 + color_seg * 0.5).astype(np.uint8)
 
     fig = draw_plot(pred_img, seg)
     return fig
 
-demo = gr.Interface(fn=sepia,
-                    inputs=gr.Image(shape=(400, 600)),
-                    outputs=['plot'],
-                    examples=["person-1.jpg", "person-2.jpg", "person-3.jpg", "person-4.jpg", "person-5.jpg"],
-                    allow_flagging='never')
-
+demo = gr.Interface(
+    fn=run_inference,
+    inputs=gr.Image(type="numpy", label="Input Image"),
+    outputs=gr.Plot(label="Overlay + Legend"),
+    examples=[
+        "ADE_val_00000001.jpeg",
+        "ADE_val_00001159.jpg",
+        "ADE_val_00001248.jpg",
+        "ADE_val_00001472.jpg"
+    ],
+    flagging_mode="never",
+    cache_examples=False,
+)
 
-demo.launch()
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -1,6 +1,6 @@
 torch
-transformers
-tensorflow
+transformers>=4.41.0
+gradio>=4.0.0
+Pillow
 numpy
-Image
 matplotlib