mal notebooks

Author: Matt Sokoloff

examples/basics/Ontology.ipynb

@@ -11,5 +11,4 @@
 3. Datasets
     - Other concepts: Tasks
 4. Ontology
-5. Labeling Front End
 6. Projects

examples/basics/README.MD

@@ -248,7 +248,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "datarow.delete(cklvfn909149e0y9n3xjbgtct)\n",
+    "datarow.delete()\n",
     "#Will remove from the dataset too"
    ]
   },

examples/basics/data_rows.ipynb

@@ -36,7 +36,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 10,
+   "execution_count": 1,
    "id": "herbal-examination",
    "metadata": {},
    "outputs": [],
@@ -55,7 +55,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 14,
+   "execution_count": 2,
    "id": "super-notification",
    "metadata": {},
    "outputs": [],
@@ -74,7 +74,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 26,
+   "execution_count": 3,
    "id": "secondary-marine",
    "metadata": {},
    "outputs": [
@@ -87,8 +87,8 @@
       "Project description: a description\n",
       "Dataset: []\n",
       "{'tools': [], 'classifications': []}\n",
-      "Reviews: <labelbox.pagination.PaginatedCollection object at 0x112163df0>\n",
-      "Benchmarks: <labelbox.pagination.PaginatedCollection object at 0x1121635b0>\n"
+      "Reviews: <labelbox.pagination.PaginatedCollection object at 0x10fb6af40>\n",
+      "Benchmarks: <labelbox.pagination.PaginatedCollection object at 0x112cac610>\n"
      ]
     }
    ],
@@ -114,7 +114,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 29,
+   "execution_count": 4,
    "id": "lined-circuit",
    "metadata": {},
    "outputs": [
@@ -134,6 +134,14 @@
     "ds.delete()"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "connected-impression",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
   {
    "cell_type": "markdown",
    "id": "statistical-mission",

examples/basics/projects.ipynb

@@ -0,0 +1,36 @@
+import cv2
+import numpy as np
+import requests
+from io import BytesIO
+from PIL import Image
+
+def visualize_bbox_upload(image, bbox_ndjsons, color):
+    for bbox_ndjson in bbox_ndjsons:
+        bbox = bbox_ndjson['bbox']
+        start_point = (int(bbox['left']), int(bbox['top']))
+        end_point = (int(bbox['left'] + bbox['width']), int(bbox['top'] + bbox['height']))
+        image = cv2.rectangle(image, start_point, end_point, thickness = 2, color = color)
+    return image
+
+def visualize_poly_upload(image, poly_ndjsons, color):
+    for poly_ndjson in poly_ndjsons:
+        pts = [[poly['x'], poly['y']] for poly in poly_ndjson['polygon']]
+        pts = np.array(pts).astype(np.int32)
+        image = cv2.polylines(image,[pts],True,color, thickness = 2)
+    return image
+
+def visualize_point_upload(image, point_ndjsons, color):
+    for point_ndjson in point_ndjsons:
+        point = point_ndjson['point']
+        image = cv2.circle(image, (point['x'],point['y']), radius=10, color=color, thickness=-1)
+    return image
+
+def visualize_mask_uploads(image, mask_ndjsons):
+    masks = []
+    for ndjson in mask_ndjsons:
+        instanceURI = ndjson['mask']['instanceURI']
+        masks.append(np.array(Image.open(BytesIO(requests.get(instanceURI).content))))
+    masks = np.sum(masks, axis = 0).astype(np.uint8)
+    return cv2.addWeighted(image, 0.7, masks, 0.3, 0)
+
+

examples/model_assisted_labeling/image_mal.ipynb

@@ -0,0 +1,120 @@
+import tensorflow as tf
+import numpy as np
+import cv2
+
+#https://colab.research.google.com/github/tensorflow/tpu/blob/master/models/official/mask_rcnn/mask_rcnn_demo.ipynb#scrollTo=2oZWLz4xXsyQ
+
+class_mappings = {1: 'person', 3 : 'car', 28: 'umbrella', 31: 'handbag'}
+session = tf.compat.v1.Session()
+
+def load_model():
+    saved_model_dir = 'gs://cloud-tpu-checkpoints/mask-rcnn/1555659850'
+    _ = tf.compat.v1.saved_model.loader.load(session, ['serve'], saved_model_dir)
+
+
+def predict(np_image_string, min_score, height, width):
+    num_detections, detection_boxes, detection_classes, detection_scores, detection_masks, image_info = session.run(
+    ['NumDetections:0', 'DetectionBoxes:0', 'DetectionClasses:0', 'DetectionScores:0', 'DetectionMasks:0', 'ImageInfo:0'],
+    feed_dict={'Placeholder:0': np_image_string})
+    num_detections = np.squeeze(num_detections.astype(np.int32), axis=(0,))
+    detection_scores = np.squeeze(detection_scores, axis=(0,))[0:num_detections]
+    response =  {
+        'boxes' : np.squeeze(detection_boxes * image_info[0, 2], axis=(0,))[0:num_detections],
+        'class_indices' : np.squeeze(detection_classes.astype(np.int32), axis=(0,))[0:num_detections],
+    }
+    ymin, xmin, ymax, xmax = np.split(response['boxes'], 4, axis=-1)
+    instance_masks = np.squeeze(detection_masks, axis=(0,))[0:num_detections]
+    processed_boxes = np.concatenate([xmin, ymin, xmax - xmin, ymax - ymin], axis=-1)
+    response.update({'seg_masks' : generate_segmentation_from_masks(instance_masks, processed_boxes, height, width)})
+    keep_indices = detection_scores > min_score
+    keep_indices = keep_indices & np.isin(response['class_indices'], list(class_mappings.keys()))
+    for key in response:
+        response[key] = response[key][keep_indices]
+    return response
+
+
+def expand_boxes(boxes, scale):
+    """Expands an array of boxes by a given scale."""
+    # Reference: https://github.com/facebookresearch/Detectron/blob/master/detectron/utils/boxes.py#L227  # pylint: disable=line-too-long
+    # The `boxes` in the reference implementation is in [x1, y1, x2, y2] form,
+    # whereas `boxes` here is in [x1, y1, w, h] form
+    w_half = boxes[:, 2] * .5
+    h_half = boxes[:, 3] * .5
+    x_c = boxes[:, 0] + w_half
+    y_c = boxes[:, 1] + h_half
+
+    w_half *= scale
+    h_half *= scale
+
+    boxes_exp = np.zeros(boxes.shape)
+    boxes_exp[:, 0] = x_c - w_half
+    boxes_exp[:, 2] = x_c + w_half
+    boxes_exp[:, 1] = y_c - h_half
+    boxes_exp[:, 3] = y_c + h_half
+
+    return boxes_exp
+
+def generate_segmentation_from_masks(masks,
+                                     detected_boxes,
+                                     image_height,
+                                     image_width,
+                                     is_image_mask=False):
+    """Generates segmentation result from instance masks.
+    Args:
+      masks: a numpy array of shape [N, mask_height, mask_width] representing the
+        instance masks w.r.t. the `detected_boxes`.
+      detected_boxes: a numpy array of shape [N, 4] representing the reference
+        bounding boxes.
+      image_height: an integer representing the height of the image.
+      image_width: an integer representing the width of the image.
+      is_image_mask: bool. True: input masks are whole-image masks. False: input
+        masks are bounding-box level masks.
+    Returns:
+      segms: a numpy array of shape [N, image_height, image_width] representing
+        the instance masks *pasted* on the image canvas.
+    """
+
+
+    _, mask_height, mask_width = masks.shape
+    scale = max((mask_width + 2.0) / mask_width,
+                (mask_height + 2.0) / mask_height)
+
+    ref_boxes = expand_boxes(detected_boxes, scale)
+    ref_boxes = ref_boxes.astype(np.int32)
+    padded_mask = np.zeros((mask_height + 2, mask_width + 2), dtype=np.float32)
+    segms = []
+    for mask_ind, mask in enumerate(masks):
+      im_mask = np.zeros((image_height, image_width), dtype=np.uint8)
+      if is_image_mask:
+        # Process whole-image masks.
+        im_mask[:, :] = mask[:, :]
+      else:
+        # Process mask inside bounding boxes.
+        padded_mask[1:-1, 1:-1] = mask[:, :]
+
+        ref_box = ref_boxes[mask_ind, :]
+        w = ref_box[2] - ref_box[0] + 1
+        h = ref_box[3] - ref_box[1] + 1
+        w = np.maximum(w, 1)
+        h = np.maximum(h, 1)
+
+        mask = cv2.resize(padded_mask, (w, h))
+        mask = np.array(mask > 0.5, dtype=np.uint8)
+
+        x_0 = max(ref_box[0], 0)
+        x_1 = min(ref_box[2] + 1, image_width)
+        y_0 = max(ref_box[1], 0)
+        y_1 = min(ref_box[3] + 1, image_height)
+
+        im_mask[y_0:y_1, x_0:x_1] = mask[(y_0 - ref_box[1]):(y_1 - ref_box[1]), (
+            x_0 - ref_box[0]):(x_1 - ref_box[0])]
+      segms.append(im_mask)
+
+    segms = np.array(segms)
+    assert masks.shape[0] == segms.shape[0]
+    return segms
+
+
+
+
+