clarifai 11.3.0rc2__py3-none-any.whl → 11.4.0__py3-none-any.whl

clarifai/utils/evaluation/testset_annotation_parser.py

@@ -1,150 +1,160 @@
 from typing import List, Tuple
+
 import numpy as np
 from clarifai_grpc.grpc.api import resources_pb2 as respb2
 
 
 def parse_eval_annotation_classifier(
-    eval_metrics: respb2.EvalMetrics) -> Tuple[np.array, np.array, list, List[respb2.Input]]:
-  test_set = eval_metrics.test_set
-  # get concept ids
-  concept_ids = [each.id for each in test_set[0].predicted_concepts]
-  concept_ids.sort()
-  # get test set
-  y_preds = []
-  y = []
-  inputs = []
-  for data in test_set:
-
-    def _to_array(_data):
-      cps = [0] * len(concept_ids)
-      for each in _data:
-        cps[concept_ids.index(each.id)] = each.value
-      return np.asarray(cps)
-
-    y_preds.append(_to_array(data.predicted_concepts))
-    y.append(_to_array(data.ground_truth_concepts))
-    inputs.append(data.input)
-
-  return np.asarray(y), np.asarray(y_preds), concept_ids, inputs
-
-
-def parse_eval_annotation_detector(eval_metrics: respb2.EvalMetrics,
-                                   normalized_box: bool = False,
-                                   box_style: str = "xyxy"
-                                  ) -> Tuple[np.array, np.array, list, List[respb2.Input]]:
-  BOX_STYLES = ["xyxy", "xywh"]
-  assert box_style in BOX_STYLES, ValueError(f"Expected box_style in {BOX_STYLES}")
-
-  concept_ids = list(set([each.concept.id for each in eval_metrics.metrics_by_class]))
-  concept_ids.sort()
-
-  def _get_box_annot(field, img_height, img_width):
-    xyxy_concept_score = []
-    for each in field:
-      box = each.region_info.bounding_box
-      x1 = box.left_col * img_width
-      y1 = box.top_row * img_height
-      x2 = box.right_col * img_width
-      y2 = box.bottom_row * img_height
-      score = each.data.concepts[0].value
-      concept = each.data.concepts[0].id
-      concept_index = concept_ids.index(concept)
-      if box_style == "xyxy":
-        xyxy_concept_score.append([x1, y1, x2, y2, concept_index, score])
-      else:
-        w = abs(x1 - x2)
-        h = abs(y1 - y2)
-        xyxy_concept_score.append([x1, y1, w, h, concept_index, score])
-
-    return np.asarray(xyxy_concept_score)
-
-  inputs = []
-  pred_xyxy_concept_score = []
-  gt_xyxy_concept_score = []
-  for input_data in eval_metrics.test_set:
-    _input = input_data.input
-    img_height = _input.data.image.image_info.height if not normalized_box else 1.
-    img_width = _input.data.image.image_info.height if not normalized_box else 1.
-    _pred_xyxy_concept_score = _get_box_annot(
-        input_data.predicted_annotation.data.regions, img_height=img_height, img_width=img_width)
-    _gt_xyxy_concept_score = _get_box_annot(
-        input_data.ground_truth_annotation.data.regions,
-        img_height=img_height,
-        img_width=img_width)
-
-    pred_xyxy_concept_score.append(_pred_xyxy_concept_score)
-    gt_xyxy_concept_score.append(_gt_xyxy_concept_score)
-    inputs.append(_input)
-
-  return np.asarray(gt_xyxy_concept_score), np.asarray(
-      pred_xyxy_concept_score), concept_ids, inputs
-
-
-def parse_eval_annotation_detector_coco(eval_metrics: respb2.EvalMetrics,
-                                       ) -> Tuple[np.array, np.array, list, List[respb2.Input]]:
-
-  gts, preds, concept_ids, inputs = parse_eval_annotation_detector(
-      eval_metrics=eval_metrics, normalized_box=False, box_style="xywh")
-
-  def _make_box_annot(data, input_data, accum_id, is_pred=True):
-    img_id = input_data.id
-    img_url = input_data.data.image.url
-    img_height = input_data.data.image.image_info.height
-    img_width = input_data.data.image.image_info.height
-    image = {
-        "id": img_id,
-        "file_name": img_url,
-        "width": img_width,
-        "height": img_height,
-    }
-    annotations = []
-    for i, d in enumerate(data.tolist()):
-      area = d[2] * d[3]
-      box = {
-          "iscrowd": 0,
-          "ignore": 0,
-          "image_id": img_id,
-          "bbox": d[:4],
-          "area": area,
-          "segmentation": [],
-          "category_id": d[4],
-          "id": accum_id + i,
-      }
-      if is_pred:
-        box["score"] = d[5]
-      annotations.append(box)
-
-    return image, annotations
-
-  categories = [{
-      "supercategory": "none",
-      "id": i,
-      "name": label
-  } for i, label in enumerate(concept_ids)]
-
-  accum_pred_ids = 0
-  accum_gt_ids = 0
-  pred_images = []
-  pred_boxes = []
-  gt_images = []
-  gt_boxes = []
-  for ith, input_proto in enumerate(inputs):
-    pred = preds[ith]
-    gt = gts[ith]
-    pred_img, pred_box = _make_box_annot(
-        pred, accum_id=accum_pred_ids, input_data=input_proto, is_pred=True)
-    gt_img, gt_box = _make_box_annot(
-        gt, accum_id=accum_gt_ids, input_data=input_proto, is_pred=False)
-
-    accum_pred_ids += len(pred)
-    pred_images.append(pred_img)
-    pred_boxes += pred_box
-
-    accum_gt_ids += len(gt)
-    gt_images.append(gt_img)
-    gt_boxes += gt_box
-
-  pred_annots = {"images": pred_images, "annotations": pred_boxes, "categories": categories}
-  gt_annots = {"images": gt_images, "annotations": gt_boxes, "categories": categories}
-
-  return gt_annots, pred_annots
+    eval_metrics: respb2.EvalMetrics,
+) -> Tuple[np.array, np.array, list, List[respb2.Input]]:
+    test_set = eval_metrics.test_set
+    # get concept ids
+    concept_ids = [each.id for each in test_set[0].predicted_concepts]
+    concept_ids.sort()
+    # get test set
+    y_preds = []
+    y = []
+    inputs = []
+    for data in test_set:
+
+        def _to_array(_data):
+            cps = [0] * len(concept_ids)
+            for each in _data:
+                cps[concept_ids.index(each.id)] = each.value
+            return np.asarray(cps)
+
+        y_preds.append(_to_array(data.predicted_concepts))
+        y.append(_to_array(data.ground_truth_concepts))
+        inputs.append(data.input)
+
+    return np.asarray(y), np.asarray(y_preds), concept_ids, inputs
+
+
+def parse_eval_annotation_detector(
+    eval_metrics: respb2.EvalMetrics, normalized_box: bool = False, box_style: str = "xyxy"
+) -> Tuple[np.array, np.array, list, List[respb2.Input]]:
+    BOX_STYLES = ["xyxy", "xywh"]
+    assert box_style in BOX_STYLES, ValueError(f"Expected box_style in {BOX_STYLES}")
+
+    concept_ids = list(set([each.concept.id for each in eval_metrics.metrics_by_class]))
+    concept_ids.sort()
+
+    def _get_box_annot(field, img_height, img_width):
+        xyxy_concept_score = []
+        for each in field:
+            box = each.region_info.bounding_box
+            x1 = box.left_col * img_width
+            y1 = box.top_row * img_height
+            x2 = box.right_col * img_width
+            y2 = box.bottom_row * img_height
+            score = each.data.concepts[0].value
+            concept = each.data.concepts[0].id
+            concept_index = concept_ids.index(concept)
+            if box_style == "xyxy":
+                xyxy_concept_score.append([x1, y1, x2, y2, concept_index, score])
+            else:
+                w = abs(x1 - x2)
+                h = abs(y1 - y2)
+                xyxy_concept_score.append([x1, y1, w, h, concept_index, score])
+
+        return np.asarray(xyxy_concept_score)
+
+    inputs = []
+    pred_xyxy_concept_score = []
+    gt_xyxy_concept_score = []
+    for input_data in eval_metrics.test_set:
+        _input = input_data.input
+        img_height = _input.data.image.image_info.height if not normalized_box else 1.0
+        img_width = _input.data.image.image_info.height if not normalized_box else 1.0
+        _pred_xyxy_concept_score = _get_box_annot(
+            input_data.predicted_annotation.data.regions,
+            img_height=img_height,
+            img_width=img_width,
+        )
+        _gt_xyxy_concept_score = _get_box_annot(
+            input_data.ground_truth_annotation.data.regions,
+            img_height=img_height,
+            img_width=img_width,
+        )
+
+        pred_xyxy_concept_score.append(_pred_xyxy_concept_score)
+        gt_xyxy_concept_score.append(_gt_xyxy_concept_score)
+        inputs.append(_input)
+
+    return (
+        np.asarray(gt_xyxy_concept_score),
+        np.asarray(pred_xyxy_concept_score),
+        concept_ids,
+        inputs,
+    )
+
+
+def parse_eval_annotation_detector_coco(
+    eval_metrics: respb2.EvalMetrics,
+) -> Tuple[np.array, np.array, list, List[respb2.Input]]:
+    gts, preds, concept_ids, inputs = parse_eval_annotation_detector(
+        eval_metrics=eval_metrics, normalized_box=False, box_style="xywh"
+    )
+
+    def _make_box_annot(data, input_data, accum_id, is_pred=True):
+        img_id = input_data.id
+        img_url = input_data.data.image.url
+        img_height = input_data.data.image.image_info.height
+        img_width = input_data.data.image.image_info.height
+        image = {
+            "id": img_id,
+            "file_name": img_url,
+            "width": img_width,
+            "height": img_height,
+        }
+        annotations = []
+        for i, d in enumerate(data.tolist()):
+            area = d[2] * d[3]
+            box = {
+                "iscrowd": 0,
+                "ignore": 0,
+                "image_id": img_id,
+                "bbox": d[:4],
+                "area": area,
+                "segmentation": [],
+                "category_id": d[4],
+                "id": accum_id + i,
+            }
+            if is_pred:
+                box["score"] = d[5]
+            annotations.append(box)
+
+        return image, annotations
+
+    categories = [
+        {"supercategory": "none", "id": i, "name": label} for i, label in enumerate(concept_ids)
+    ]
+
+    accum_pred_ids = 0
+    accum_gt_ids = 0
+    pred_images = []
+    pred_boxes = []
+    gt_images = []
+    gt_boxes = []
+    for ith, input_proto in enumerate(inputs):
+        pred = preds[ith]
+        gt = gts[ith]
+        pred_img, pred_box = _make_box_annot(
+            pred, accum_id=accum_pred_ids, input_data=input_proto, is_pred=True
+        )
+        gt_img, gt_box = _make_box_annot(
+            gt, accum_id=accum_gt_ids, input_data=input_proto, is_pred=False
+        )
+
+        accum_pred_ids += len(pred)
+        pred_images.append(pred_img)
+        pred_boxes += pred_box
+
+        accum_gt_ids += len(gt)
+        gt_images.append(gt_img)
+        gt_boxes += gt_box
+
+    pred_annots = {"images": pred_images, "annotations": pred_boxes, "categories": categories}
+    gt_annots = {"images": gt_images, "annotations": gt_boxes, "categories": categories}
+
+    return gt_annots, pred_annots