Uploading Trashify box detection model app.py

.gitattributes

@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+trashify_examples/trashify_example_2.jpeg filter=lfs diff=lfs merge=lfs -text
+trashify_examples/trashify_example_3.jpeg filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,12 +1,36 @@
 ---
-title: Trashify Demo
-emoji: 🌍
-colorFrom: blue
-colorTo: yellow
+title: Trashify Demo V4 🚮
+emoji: 🗑️
+colorFrom: purple
+colorTo: blue
 sdk: gradio
-sdk_version: 6.4.0
+sdk_version: 5.34.0
 app_file: app.py
 pinned: false
+license: apache-2.0
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# 🚮 Trashify Object Detector V4 
+
+Object detection demo to detect `trash`, `bin`, `hand`, `trash_arm`, `not_trash`, `not_bin`, `not_hand`. 
+
+Used as example for encouraging people to cleanup their local area.
+
+If `trash`, `hand`, `bin` all detected = +1 point.
+
+## Dataset
+
+All Trashify models are trained on a custom hand-labelled dataset of people picking up trash and placing it in a bin.
+
+The dataset can be found on Hugging Face as [`mrdbourke/trashify_manual_labelled_images`](https://huggingface.co/datasets/mrdbourke/trashify_manual_labelled_images).
+
+## Demos
+
+* [V1](https://huggingface.co/spaces/mrdbourke/trashify_demo_v1) = Fine-tuned [Conditional DETR](https://huggingface.co/docs/transformers/en/model_doc/conditional_detr) model trained *without* data augmentation.
+* [V2](https://huggingface.co/spaces/mrdbourke/trashify_demo_v2) = Fine-tuned Conditional DETR model trained *with* data augmentation.
+* [V3](https://huggingface.co/spaces/mrdbourke/trashify_demo_v3) = Fine-tuned Conditional DETR model trained *with* data augmentation (same as V2) with an NMS (Non Maximum Suppression) post-processing step.
+* [V4](https://huggingface.co/spaces/mrdbourke/trashify_demo_v4) = Fine-tuned [RT-DETRv2](https://huggingface.co/docs/transformers/main/en/model_doc/rt_detr_v2) model trained *without* data augmentation or NMS post-processing (current best mAP).
+
+## Learn more
+
+See the full end-to-end code of how this demo was built at [learnhuggingface.com](https://www.learnhuggingface.com/notebooks/hugging_face_object_detection_tutorial). 

app.py

@@ -0,0 +1,169 @@
+
+# 1. Import the required libraries and packages
+import gradio as gr
+import torch
+from PIL import Image, ImageDraw, ImageFont # could also use torch utilities for drawing
+
+from transformers import AutoImageProcessor
+from transformers import AutoModelForObjectDetection
+
+### 2. Setup preprocessing and helper functions ###
+
+# Setup target model path to load
+# Note: Can load from Hugging Face or can load from local 
+model_save_path = "mrdbourke/rt_detrv2_finetuned_trashify_box_detector_v1"
+
+# Load the model and preprocessor
+# Because this app.py file is running directly on Hugging Face Spaces, the model will be loaded from the Hugging Face Hub
+image_processor = AutoImageProcessor.from_pretrained(model_save_path)
+model = AutoModelForObjectDetection.from_pretrained(model_save_path)
+
+# Set the target device (use CUDA/GPU if it is available)
+device = "cuda" if torch.cuda.is_available() else "cpu"
+model = model.to(device)
+
+# Get the id2label dictionary from the model
+id2label = model.config.id2label
+
+# Set up a colour dictionary for plotting boxes with different colours
+color_dict = {   
+    "bin": "green",
+    "trash": "blue",
+    "hand": "purple",
+    "trash_arm": "yellow",
+    "not_trash": "red",
+    "not_bin": "red",
+    "not_hand": "red",
+}
+
+### 3. Create function to predict on a given image with a given confidence threshold ###
+def predict_on_image(image, conf_threshold):
+    # Make sure model is in eval mode
+    model.eval()
+
+    # Make a prediction on target image 
+    with torch.no_grad():
+        inputs = image_processor(images=[image], return_tensors="pt")
+        model_outputs = model(**inputs.to(device))
+
+        target_sizes = torch.tensor([[image.size[1], image.size[0]]]) # -> [batch_size, height, width] 
+
+        # Post process the raw outputs from the model 
+        results = image_processor.post_process_object_detection(model_outputs,
+                                                                threshold=conf_threshold,
+                                                                target_sizes=target_sizes)[0]
+
+    # Return all items in results to CPU (we'll want this for displaying outputs with matplotlib)
+    for key, value in results.items():
+        try:
+            results[key] = value.item().cpu() # can't get scalar as .item() so add try/except block
+        except:
+            results[key] = value.cpu()
+
+    ### 4. Draw the predictions on the target image ###
+
+    # Can return results as plotted on a PIL image (then display the image)
+    draw = ImageDraw.Draw(image)
+
+    # Get a font from ImageFont
+    font = ImageFont.load_default(size=20)
+
+    # Get class names as text for print out
+    detected_class_name_text_labels = []
+
+    # Iterate through the predictions of the model and draw them on the target image
+    for box, score, label in zip(results["boxes"], results["scores"], results["labels"]):
+        # Create coordinates
+        x, y, x2, y2 = tuple(box.tolist())
+
+        # Get label_name
+        label_name = id2label[label.item()]
+        targ_color = color_dict[label_name]
+        detected_class_name_text_labels.append(label_name)
+
+        # Draw the rectangle
+        draw.rectangle(xy=(x, y, x2, y2), 
+                       outline=targ_color,
+                       width=3)
+
+        # Create a text string to display
+        text_string_to_show = f"{label_name} ({round(score.item(), 3)})"
+
+        # Draw the text on the image
+        draw.text(xy=(x, y),
+                  text=text_string_to_show,
+                  fill="white",
+                  font=font)
+
+    # Remove the draw each time
+    del draw
+
+    ### 5. Create logic for outputting information message ### 
+
+    # Setup set of target items to discover
+    target_items = {"trash", "bin", "hand"} 
+    detected_items = set(detected_class_name_text_labels)
+
+    # If no items detected or trash, bin, hand not in detected items, return notification
+    if not detected_items & target_items:
+        return_string = (
+            f"No trash, bin or hand detected at confidence threshold {conf_threshold}. "
+            "Try another image or lowering the confidence threshold."
+        )
+        print(return_string)
+        return image, return_string
+
+    # If there are missing items, say what the missing items are
+    missing_items = target_items - detected_items
+    if missing_items:
+        return_string = (
+            f"Detected the following items: {sorted(detected_items & target_items)}. But missing the following in order to get +1: {sorted(missing_items)}. "
+            "If this is an error, try another image or altering the confidence threshold. "
+            "Otherwise, the model may need to be updated with better data."
+        )
+        print(return_string)
+        return image, return_string
+
+    # If all target items are present (the final remaining case)
+    return_string = f"+1! Found the following items: {sorted(detected_items)}, thank you for cleaning up the area!"
+    print(return_string)
+    return image, return_string
+
+### 6. Setup the demo application to take in image, make a prediction with our model, return the image with drawn predicitons ### 
+
+# Write description for our demo application
+description = """
+Help clean up your local area! Upload an image and get +1 if there is all of the following items detected: trash, bin, hand.
+
+Model is a fine-tuned version of [RT-DETRv2](https://huggingface.co/docs/transformers/main/en/model_doc/rt_detr_v2#transformers.RTDetrV2Config) on the [Trashify dataset](https://huggingface.co/datasets/mrdbourke/trashify_manual_labelled_images).
+
+See the full data loading and training code on [learnhuggingface.com](https://www.learnhuggingface.com/notebooks/hugging_face_object_detection_tutorial).
+
+This version is v4 because the first three versions were using a different model and did not perform as well, see the [README](https://huggingface.co/spaces/mrdbourke/trashify_demo_v4/blob/main/README.md) for more.
+"""
+
+# Create the Gradio interface to accept an image and confidence threshold and return an image with drawn prediction boxes
+demo = gr.Interface(
+    fn=predict_on_image,
+    inputs=[
+        gr.Image(type="pil", label="Target Image"),
+        gr.Slider(minimum=0, maximum=1, value=0.3, label="Confidence Threshold")
+    ],
+    outputs=[
+        gr.Image(type="pil", label="Image Output"),
+        gr.Text(label="Text Output")
+    ],
+    title="🚮 Trashify Object Detection Demo V4",
+    description=description,
+    # Examples come in the form of a list of lists, where each inner list contains elements to prefill the `inputs` parameter with
+    # See where the examples originate from here: https://huggingface.co/datasets/mrdbourke/trashify_examples/
+    examples=[
+        ["trashify_examples/trashify_example_1.jpeg", 0.3],
+        ["trashify_examples/trashify_example_2.jpeg", 0.3], 
+        ["trashify_examples/trashify_example_3.jpeg", 0.3],
+    ],
+    cache_examples=True
+)
+
+# Launch the demo
+demo.launch()

requirements.txt

@@ -0,0 +1,4 @@
+timm
+gradio
+torch
+transformers