HMSI - analyzer output fix

docs/tutorial/3/3b.yaml

@@ -3,6 +3,8 @@ elements_in:
   - Youtube
 analyse:
   - name: Frames
+    config:
+      in_parallel: false
   - name: KerasPretrained
     config:
       in_parallel: false

docs/tutorial/3/3c.yaml

@@ -1,20 +1,29 @@
-folder: media/demo_official/3c
+#Where on our computer do we want to store the media?
+folder: media/teargas_mexico_2022
 select:
+  #Which media platforms do we want to search?
   name: Youtube
   config:
+    #What are our search parameters?
     search_term: tear gas + mexico
-    uploaded_before: "2018-11-30T00:00:00Z"
-    uploaded_after: "2018-11-15T00:00:00Z"
+    uploaded_before: "2022-11-30T00:00:00Z"
+    uploaded_after: "2022-11-20T00:00:00Z"
 analyse:
+  #Once we have a video, we process it by splitting it into images, 
+  #one for each frame
   - name: Frames
+  #We don't want to analyze successive frames that are very similar,
+  # so we deduplicate these
   - name: ImageDedup
     config:
       threshold: 3
       method: dhash
+  #We're now ready to run the image classification process over each frame.
+  #In this example, we'll use Resnet, an image classification model pretrained 
+  # on the publicly available Imagenet database
   - name: KerasPretrained
     config:
       model: ResNet50
+      #We're interested in detecting anything with the "tear gas" label
       labels:
-        - tank
-        - rifle
-        - military uniform
+        - tear gas

docs/tutorial/3/3d.yaml

@@ -0,0 +1,19 @@
+folder: media/demo_official/3d
+select:
+  name: Local
+  config:
+    source: media/demo_official/3c/youtube/derived/Frames
+analyse:
+  - name: ImageDedup
+    config:
+      threshold: 3
+      method: dhash
+  - name: KerasPretrained
+    config:
+      model: ResNet50
+      labels:
+        - tear gas canister
+        - tear gas
+        - tank
+        - rifle
+        - military uniform

results.py

@@ -0,0 +1,79 @@
+import os
+import json
+import sys
+
+def get_video_meta(video_id, keras_pretrained_path):
+    """Fetch the video title, upload date, and view count from the meta.json file."""
+    meta_file_path = os.path.join(keras_pretrained_path, 'data', video_id, 'meta.json')
+    with open(meta_file_path, 'r') as meta_file:
+        meta_data = json.load(meta_file)
+    title = meta_data.get('title', 'N/A')
+    upload_date = meta_data.get('upload_date', 'N/A')
+    view_count = meta_data.get('view_count', 'N/A')
+    return title, upload_date, view_count
+
+def main(keras_pretrained_path):
+    # Dictionary to store the results
+    labels_dict = {}
+
+    # Iterate through the folders inside the 'derived' directory
+    for video_id in os.listdir(keras_pretrained_path):
+        video_path = os.path.join(keras_pretrained_path, video_id)
+
+        # Check if it's a directory (i.e., a video ID folder)
+        if os.path.isdir(video_path):
+            # Path to the preds.json file for the current video
+            preds_file_path = os.path.join(video_path, 'preds.json')
+
+            # Check if the preds.json file exists
+            if os.path.exists(preds_file_path):
+                with open(preds_file_path, 'r') as file:
+                    content = file.read()
+
+                    # Check if the file is empty
+                    if content:
+                        # Parse the JSON content
+                        preds_data = json.loads(content)
+
+                        # Iterate through the labels and their data
+                        for label, data in preds_data.get('labels', {}).items():
+                            frames = data.get('frames', [])
+                            scores = data.get('scores', [])
+
+                            # Compile the frames and their corresponding scores
+                            frame_scores = [{"frame": frame, "score": score} for frame, score in zip(frames, scores)]
+
+                            # Get video metadata
+                            title, upload_date, view_count = get_video_meta(video_id, keras_pretrained_path)
+
+                            # Append the video info to the label's list
+                            if label not in labels_dict:
+                                labels_dict[label] = []
+                            labels_dict[label].append({
+                                "video_id": video_id,
+                                "video_title": title,
+                                "upload_date": upload_date,
+                                "view_count": view_count,
+                                "frame_scores": frame_scores
+                            })
+
+    # Print the results
+    for label, videos in labels_dict.items():
+        print(f"Assigned Label: {label}")
+        print(f"Total Number of Videos with the Label Found: {len(videos)}")
+        print("Videos where label was found:")
+        for video in videos:
+            print(f"  Video Title: {video['video_title']} (ID: {video['video_id']})")
+            print(f"  Upload Date: {video['upload_date']}")
+            print(f"  View Count: {video['view_count']}")
+            print("  Frames:")
+            for frame_score in video['frame_scores']:
+                print(f"    Frame: {frame_score['frame']}, Score: {frame_score['score']}")
+            print()
+
+if __name__ == "__main__":
+    if len(sys.argv) < 2:
+        print("Please provide the path to the Keras_Pretrained directory")
+        sys.exit(1)
+    keras_pretrained_path = sys.argv[1]
+    main(keras_pretrained_path)

resumer.yaml

@@ -0,0 +1,19 @@
+folder: media/teargas_mexico_2022
+select:
+  name: Local
+  config:
+    source: media/teargas_mexico_2022/Youtube/derived/Frames
+analyse:
+  - name: ImageDedup
+    config:
+      threshold: 3
+      method: dhash
+  #We're now ready to run the image classification process over each frame.
+  #In this example, we'll use Resnet, an image classification model pretrained 
+  # on the publicly available Imagenet database
+  - name: KerasPretrained
+    config:
+      model: ResNet50
+      #We're interested in detecting anything with the "tear gas" label
+      labels:
+        - tear gas

src/lib/common/analyser.py

@@ -87,7 +87,7 @@ def start_analysing(self):
             self.disk.write_meta(
                 f"{self.get_selector()}/{self.name}",
                 {
-                    "etype": self.out_etype.__repr__(),
+                    "etype": "CvJson",
                     "config": cfg,
                     "stage": {"name": self.name, "module": "analyser"},
                 },

src/lib/selectors/Youtube/core.py

@@ -1,4 +1,4 @@
-import yt_dlp 
+import yt_dlp
 import json
 import re
 import argparse, os, sys

teargas 2.yaml

@@ -0,0 +1,34 @@
+#Where on our computer do we want to store the media?
+folder: media/teargas_mexico_2018
+select:
+	#Which media platforms do we want to search?
+  name: Youtube
+	config:
+		#What are our search parameters?
+ 		search_term: tear gas + mexico
+		uploaded_after: "2018-11-25T00:00:00Z"
+		uploaded_before: "2018-11-26T12:00:00Z"
+analyse:
+#Once we have a video, we process it by splitting it into images, one for each frame
+	- name: Frames
+#We don't want to analyze successive frames that are very similar, so we deduplicate these
+	- name: ImageDedup
+	config:
+		threshold: 3
+		method: dhash
+#We're now ready to run the image classification process over each frame.
+#In this example, we'll use Resnet, an image classification model pretrained on the publicly available Imagenet database
+	- name: KerasPretrained
+	config:
+		model: ResNet50
+		labels:
+		# We're looking for tear gas and other indicators of police force
+			- tear gas
+			- rifle
+			- police
+			- smoke
+			- military uniform
+
+
+
+

ukraine.yaml

@@ -0,0 +1,28 @@
+# Where on our computer do we want to store the media?
+folder: media/tanks_in_ukraine_july2014
+
+# Which media platforms do we want to search?
+select:
+  name: Youtube
+  config:
+    # What are our search parameters?
+    search_term: tanks + ukraine
+    uploaded_after: "2014-07-20T00:00:00Z"
+    uploaded_before: "2014-07-30T00:00:00Z"
+
+analyse:
+  # Once we have a video, we process it by splitting it into images, one for each frame. 
+  - name: Frames
+  # We don't want to analyze successive frames that are very similar, so we deduplicate these
+  - name: ImageDedup
+    config:
+      threshold: 3
+      method: dhash
+  # We're now ready to run the image classification algorithm over each frame.
+  # In this example, we'll use Resnet, a model pretrained on the publicly available Imagenet database.
+  - name: KerasPretrained
+    config:
+      model: ResNet50
+      # We're interested in detecting anything with the label "tank"
+      labels: 
+        - "tank"

vidtester.yaml

@@ -0,0 +1,18 @@
+folder: media/tanks_in_ukraine_july2014
+select:
+  name: Youtube
+  config:
+    search_term: "tanks + ukraine"
+    uploaded_before: "2014-7-25T00:00:00Z"
+    uploaded_after: "2014-7-20T00:00:00Z"
+analyse:
+  - name: Frames
+  - name: ImageDedup
+    config:
+      threshold: 3
+      method: dhash
+  - name: KerasPretrained
+    config:
+      model: ResNet50
+      labels:
+        - tank