make more gradionic

README.md

@@ -1,17 +1,33 @@
 ---
-title: SegFormer (ADE20k) in TensorFlow
+title: SegFormer (ADE20k) in ONNX
 emoji: 🏃
 colorFrom: indigo
 colorTo: gray
 sdk: gradio
-sdk_version: 5.47.1
+sdk_version: 5.49.0
 app_file: app.py
 pinned: false
 license: apache-2.0
+short_description: Image segmentation fine-tuned on the ADE20k dataset
 ---
 
+## SegFormer (ADE20k) in ONNX
+
+This is demo TensorFlow SegFormer from 🤗 `transformers` official package. The pre-trained model was trained to segment scene specific images. We are **currently using ONNX model converted from the TensorFlow based SegFormer to improve the latency**. The average latency of an inference is **21** and **8** seconds for TensorFlow and ONNX converted models respectively (in [Colab](https://github.com/deep-diver/segformer-tf-transformers/blob/main/notebooks/TFSegFormer_ONNX.ipynb)). Check out the [repository](https://github.com/deep-diver/segformer-tf-transformers) to find out how to make inference, finetune the model with custom dataset, and further information.
+
+
 This space hosts the [SegFormer model](https://arxiv.org/abs/2105.15203) in TensorFlow. This model was fine-tuned on the [ADE20k dataset](http://groups.csail.mit.edu/vision/datasets/ADE20K/). To know more about the checkpoint used in this space, refer to the model card
 [here](https://huggingface.co/nvidia/segformer-b5-finetuned-ade-640-640).
 
 Please note that since the model was fine-tuned on the ADE20k dataset, the model is expected to provide best results for images
-belonging to scene categories. For an overview of the dataset, refer to its [homepage](http://groups.csail.mit.edu/vision/datasets/ADE20K/).
+belonging to scene categories. For an overview of the dataset, refer to its [homepage](http://groups.csail.mit.edu/vision/datasets/ADE20K/).
+
+## Acknowledgments
+This work integrates code and concepts from several repositories.  
+For proper attribution, please refer to the following sources (or notify us if any are missing):
+- [Original space](https://huggingface.co/spaces/chansung/segformer-tf-transformers)
+- 
+## Contact
+For questions, comments, or feedback, please contact:  
+📧 **leonelhs@gmail.com**
+

app.py

@@ -1,13 +1,46 @@
+#######################################################################################
+#
+# MIT License
+#
+# Copyright (c) [2025] [leonelhs@gmail.com]
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+#
+#######################################################################################
+#
+# This project is one of several repositories exploring image segmentation techniques.
+# All related projects and interactive demos can be found at:
+# https://huggingface.co/spaces/leonelhs/removators
+# huggingface: https://huggingface.co/spaces/leonelhs/segformer-tf-transformers
+#
+
 import csv
 import os
 import sys
+from itertools import islice
 
 import cv2
-import gradio as gr
-import matplotlib.pyplot as plt
 import numpy as np
 import onnxruntime as ort
-from matplotlib import gridspec
+import gradio as gr
+from PIL import Image
+from huggingface_hub import hf_hub_download
 
 ade_palette = []
 labels_list = []
@@ -23,14 +56,16 @@ with open(r"ade_palette.txt", "r") as fp:
         tmp_list = list(map(int, line[:-1].strip("][").split(", ")))
         ade_palette.append(tmp_list)
 
-colormap = np.asarray(ade_palette)
+    colormap = np.asarray(ade_palette)
+
+
+REPO_ID = "leonelhs/segmentators"
+model_path = hf_hub_download(repo_id=REPO_ID, filename="segformer/segformer-b5-finetuned-ade-640-640.onnx")
+
 
-model_filename = "segformer-b5-finetuned-ade-640-640.onnx"
 sess_options = ort.SessionOptions()
 sess_options.intra_op_num_threads = os.cpu_count()
-sess = ort.InferenceSession(
-    model_filename, sess_options, providers=["CPUExecutionProvider"]
-)
+sess = ort.InferenceSession(model_path, sess_options, providers=["CPUExecutionProvider"])
 
 
 def label_to_color_image(label):
@@ -42,31 +77,8 @@ def label_to_color_image(label):
 
     return colormap[label]
 
+def predict(input_img):
 
-def draw_plot(pred_img, seg):
-    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)
-    ax = plt.subplot(grid_spec[1])
-    plt.imshow(FULL_COLOR_MAP[unique_labels].astype(np.uint8), interpolation="nearest")
-    ax.yaxis.tick_right()
-    plt.yticks(range(len(unique_labels)), LABEL_NAMES[unique_labels])
-    plt.xticks([], [])
-    ax.tick_params(width=0.0, labelsize=25)
-    return fig
-
-
-def sepia(input_img):
     img = cv2.imread(input_img)
     img = cv2.resize(img, (640, 640)).astype(np.float32)
     img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
@@ -76,42 +88,40 @@ def sepia(input_img):
     logits = sess.run(None, {"pixel_values": img_batch})[0]
 
     logits = np.transpose(logits, (0, 2, 3, 1))
-    seg = np.argmax(logits, axis=-1)[0].astype("float32")
-    seg = cv2.resize(seg, (640, 640)).astype("uint8")
-
-    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 == label, :] = color
+    segmented_mask = np.argmax(logits, axis=-1)[0].astype("float32")
+    segmented_mask = cv2.resize(segmented_mask, (640, 640)).astype("uint8")
 
-    # Convert to BGR
-    color_seg = color_seg[..., ::-1]
 
-    # Show image + mask
-    pred_img = img * 0.5 + color_seg * 0.5
-    pred_img = pred_img.astype(np.uint8)
+    parts = []
+    unique_labels = np.unique(segmented_mask)
+    label_names = np.asarray(labels_list)
 
-    fig = draw_plot(pred_img, seg)
-    return fig
+    for label in unique_labels:
+        part = np.where(segmented_mask == label)
+        color_seg = np.full((640, 640, 3), 0, dtype=np.uint8)
+        color_seg[part[0], part[1], :] = colormap[label]
+        color_seg = cv2.cvtColor(color_seg, cv2.COLOR_BGR2GRAY)
+        parts.append((color_seg, label_names[label]))
 
+    return Image.fromarray(img.astype("uint8")), parts
 
-title = "SegFormer(ADE20k) in TensorFlow"
-description = """
+with gr.Blocks(title="SegFormer") as app:
+    navbar = gr.Navbar(visible=True, main_page_name="Workspace")
+    gr.Markdown("## SegFormer(ADE20k) ONNX")
+    with gr.Row():
+        with gr.Column(scale=1):
+            inp = gr.Image(type="filepath", label="Upload Image")
+            btn_predict = gr.Button("Parse")
+        with gr.Column(scale=2):
+            out = gr.AnnotatedImage(label="Image parsed annotated")
 
-This is demo TensorFlow SegFormer from 🤗 `transformers` official package. The pre-trained model was trained to segment scene specific images. We are **currently using ONNX model converted from the TensorFlow based SegFormer to improve the latency**. The average latency of an inference is **21** and **8** seconds for TensorFlow and ONNX converted models respectively (in [Colab](https://github.com/deep-diver/segformer-tf-transformers/blob/main/notebooks/TFSegFormer_ONNX.ipynb)). Check out the [repository](https://github.com/deep-diver/segformer-tf-transformers) to find out how to make inference, finetune the model with custom dataset, and further information.
+    btn_predict.click(predict, inputs=[inp], outputs=[out])
 
-"""
 
-demo = gr.Interface(
-    sepia,
-    gr.Image(type="filepath"),
-    outputs=["plot"],
-    examples=["ADE_val_00000001.jpeg"],
-    allow_flagging="never",
-    title=title,
-    description=description,
-)
+with app.route("About this", "/about"):
+    with open("README.md") as f:
+        for line in islice(f, 12, None):
+            gr.Markdown(line.strip())
 
-demo.launch()
+app.launch(share=False, debug=True, show_error=True, mcp_server=True, pwa=True)
+app.queue()

segformer-b5-finetuned-ade-640-640.onnx

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:f27cde5816910ae5a654fed8be58e7ad0c361244b5494df18ecf37d35d7ea33c
-size 341726174