add evaluation framework and imagenet evaluation (#69)

fengyuentau · web-flow · commit 004a62710e75 · 2022-07-08T14:45:42.000+08:00
diff --git a/benchmark/config/image_classification_mobilenetv1.yaml b/benchmark/config/image_classification_mobilenetv1.yaml
@@ -17,4 +17,4 @@ Benchmark:
 Model:
   name: "MobileNetV1"
   modelPath: "models/image_classification_mobilenet/image_classification_mobilenetv1_2022apr.onnx"
-  labelPath: "models/image_classification_mobilenet/imagenet_labels.txt"
+
diff --git a/benchmark/config/image_classification_mobilenetv2.yaml b/benchmark/config/image_classification_mobilenetv2.yaml
@@ -17,4 +17,4 @@ Benchmark:
 Model:
   name: "MobileNetV2"
   modelPath: "models/image_classification_mobilenet/image_classification_mobilenetv2_2022apr.onnx"
-  labelPath: "models/image_classification_mobilenet/imagenet_labels.txt"
+
diff --git a/benchmark/config/image_classification_ppresnet.yaml b/benchmark/config/image_classification_ppresnet.yaml
@@ -17,4 +17,4 @@ Benchmark:
 Model:
   name: "PPResNet"
   modelPath: "models/image_classification_ppresnet/image_classification_ppresnet50_2022jan.onnx"
-  labelPath: "models/image_classification_ppresnet/imagenet_labels.txt"
+
diff --git a/models/image_classification_mobilenet/README.md b/models/image_classification_mobilenet/README.md
@@ -4,6 +4,17 @@ MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applicatio
 
 MobileNetV2: Inverted Residuals and Linear Bottlenecks
 
+Results of accuracy evaluation with [tools/eval](../../tools/eval).
+
+| Models | Top-1 Accuracy | Top-5 Accuracy |
+| ------ | -------------- | -------------- |
+| MobileNet V1 | 67.64 | 87.97 |
+| MobileNet V1 quant | 55.53 | 78.74 |
+| MobileNet V2 | 69.44 | 89.23 |
+| MobileNet V2 quant | 68.37 | 88.56 |
+
+\*: 'quant' stands for 'quantized'.
+
 ## Demo
 
 Run the following command to try the demo:
@@ -24,3 +35,4 @@ All files in this directory are licensed under [Apache 2.0 License](./LICENSE).
 - MobileNet V2: https://arxiv.org/abs/1801.04381
 - MobileNet V1 weight and scripts for training: https://github.com/wjc852456/pytorch-mobilenet-v1
 - MobileNet V2 weight: https://github.com/onnx/models/tree/main/vision/classification/mobilenet
+
diff --git a/models/image_classification_mobilenet/mobilenet_v1.py b/models/image_classification_mobilenet/mobilenet_v1.py
@@ -2,9 +2,11 @@
 import cv2 as cv
 
 class MobileNetV1:
-    def __init__(self, modelPath, labelPath, backendId=0, targetId=0):
+    def __init__(self, modelPath, labelPath=None, topK=1, backendId=0, targetId=0):
         self.model_path = modelPath
         self.label_path = labelPath
+        assert topK >= 1
+        self.top_k = topK
         self.backend_id = backendId
         self.target_id = targetId
 
@@ -23,9 +25,10 @@ def __init__(self, modelPath, labelPath, backendId=0, targetId=0):
 
     def _load_labels(self):
         labels = []
-        with open(self.label_path, 'r') as f:
-            for line in f:
-                labels.append(line.strip())
+        if self.label_path is not None:
+            with open(self.label_path, 'r') as f:
+                for line in f:
+                    labels.append(line.strip())
         return labels
 
     @property
@@ -61,9 +64,18 @@ def infer(self, image):
         return results
 
     def _postprocess(self, output_blob):
-        predicted_labels = []
+        batched_class_id_list = []
         for o in output_blob:
-            class_id = np.argmax(o)
-            predicted_labels.append(self.labels[class_id])
-        return predicted_labels
+            class_id_list = o.argsort()[::-1][:self.top_k]
+            batched_class_id_list.append(class_id_list)
+        if len(self.labels) > 0:
+            batched_predicted_labels = []
+            for class_id_list in batched_class_id_list:
+                predicted_labels = []
+                for class_id in class_id_list:
+                    predicted_labels.append(self._labels[class_id])
+                batched_predicted_labels.append(predicted_labels)
+            return batched_predicted_labels
+        else:
+            return batched_class_id_list
 
diff --git a/models/image_classification_mobilenet/mobilenet_v2.py b/models/image_classification_mobilenet/mobilenet_v2.py
@@ -2,9 +2,11 @@
 import cv2 as cv
 
 class MobileNetV2:
-    def __init__(self, modelPath, labelPath, backendId=0, targetId=0):
+    def __init__(self, modelPath, labelPath=None, topK=1, backendId=0, targetId=0):
         self.model_path = modelPath
         self.label_path = labelPath
+        assert topK >= 1
+        self.top_k = topK
         self.backend_id = backendId
         self.target_id = targetId
 
@@ -23,9 +25,10 @@ def __init__(self, modelPath, labelPath, backendId=0, targetId=0):
 
     def _load_labels(self):
         labels = []
-        with open(self.label_path, 'r') as f:
-            for line in f:
-                labels.append(line.strip())
+        if self.label_path is not None:
+            with open(self.label_path, 'r') as f:
+                for line in f:
+                    labels.append(line.strip())
         return labels
 
     @property
@@ -61,9 +64,18 @@ def infer(self, image):
         return results
 
     def _postprocess(self, output_blob):
-        predicted_labels = []
+        batched_class_id_list = []
         for o in output_blob:
-            class_id = np.argmax(o)
-            predicted_labels.append(self.labels[class_id])
-        return predicted_labels
+            class_id_list = o.argsort()[::-1][:self.top_k]
+            batched_class_id_list.append(class_id_list)
+        if len(self.labels) > 0:
+            batched_predicted_labels = []
+            for class_id_list in batched_class_id_list:
+                predicted_labels = []
+                for class_id in class_id_list:
+                    predicted_labels.append(self._labels[class_id])
+                batched_predicted_labels.append(predicted_labels)
+            return batched_predicted_labels
+        else:
+            return batched_class_id_list
 
diff --git a/models/image_classification_ppresnet/README.md b/models/image_classification_ppresnet/README.md
@@ -4,6 +4,15 @@ Deep Residual Learning for Image Recognition
 
 This model is ported from [PaddleHub](https://github.com/PaddlePaddle/PaddleHub) using [this script from OpenCV](https://github.com/opencv/opencv/blob/master/samples/dnn/dnn_model_runner/dnn_conversion/paddlepaddle/paddle_resnet50.py).
 
+Results of accuracy evaluation with [tools/eval](../../tools/eval).
+
+| Models | Top-1 Accuracy | Top-5 Accuracy |
+| ------ | -------------- | -------------- |
+| PP-ResNet | 82.28 | 96.15 |
+| PP-ResNet quant | 0.22 | 0.96 |
+
+\*: 'quant' stands for 'quantized'.
+
 ## Demo
 
 Run the following command to try the demo:
@@ -19,4 +28,5 @@ All files in this directory are licensed under [Apache 2.0 License](./LICENSE).
 
 - https://arxiv.org/abs/1512.03385
 - https://github.com/opencv/opencv/tree/master/samples/dnn/dnn_model_runner/dnn_conversion/paddlepaddle
-- https://github.com/PaddlePaddle/PaddleHub
+- https://github.com/PaddlePaddle/PaddleHub
+
diff --git a/models/image_classification_ppresnet/ppresnet.py b/models/image_classification_ppresnet/ppresnet.py
@@ -9,9 +9,11 @@
 import cv2 as cv
 
 class PPResNet:
-    def __init__(self, modelPath, labelPath, backendId=0, targetId=0):
+    def __init__(self, modelPath, labelPath=None, topK=1, backendId=0, targetId=0):
         self._modelPath = modelPath
         self._labelPath = labelPath
+        assert topK >= 1
+        self._topK = topK
         self._backendId = backendId
         self._targetId = targetId
 
@@ -30,9 +32,10 @@ def __init__(self, modelPath, labelPath, backendId=0, targetId=0):
 
     def _load_labels(self):
         labels = []
-        with open(self._labelPath, 'r') as f:
-            for line in f:
-                labels.append(line.strip())
+        if self._labelPath is not None:
+            with open(self._labelPath, 'r') as f:
+                for line in f:
+                    labels.append(line.strip())
         return labels
 
     @property
@@ -65,11 +68,23 @@ def infer(self, image):
         outputBlob = self._model.forward(self._outputNames)
 
         # Postprocess
-        results = self._postprocess(outputBlob)
+        results = self._postprocess(outputBlob[0])
 
         return results
 
     def _postprocess(self, outputBlob):
-        class_id = np.argmax(outputBlob[0])
-        return self._labels[class_id]
+        batched_class_id_list = []
+        for ob in outputBlob:
+            class_id_list = ob.argsort()[::-1][:self._topK]
+            batched_class_id_list.append(class_id_list)
+        if len(self._labels) > 0:
+            batched_predicted_labels = []
+            for class_id_list in batched_class_id_list:
+                predicted_labels = []
+                for class_id in class_id_list:
+                    predicted_labels.append(self._labels[class_id])
+                batched_predicted_labels.append(predicted_labels)
+            return batched_predicted_labels
+        else:
+            return batched_class_id_list
 
diff --git a/tools/eval/README.md b/tools/eval/README.md
@@ -0,0 +1,55 @@
+# Accuracy evaluation of models in OpenCV Zoo
+
+Make sure you have the following packages installed:
+
+```shell
+pip install tqdm
+```
+
+Generally speaking, evaluation can be done with the following command:
+
+```shell
+python eval.py -m model_name -d dataset_name -dr dataset_root_dir
+```
+
+Supported datasets:
+- [ImageNet](./datasets/imagenet.py)
+
+## ImageNet
+
+### Prepare data
+
+Please visit https://image-net.org/ to download the ImageNet dataset and [the labels from caffe](http://dl.caffe.berkeleyvision.org/caffe_ilsvrc12.tar.gz). Organize files as follow:
+
+```shell
+$ tree -L 2 /path/to/imagenet
+.
+├── caffe_ilsvrc12
+│   ├── det_synset_words.txt
+│   ├── imagenet.bet.pickle
+│   ├── imagenet_mean.binaryproto
+│   ├── synsets.txt
+│   ├── synset_words.txt
+│   ├── test.txt
+│   ├── train.txt
+│   └── val.txt
+├── caffe_ilsvrc12.tar.gz
+├── ILSVRC
+│   ├── Annotations
+│   ├── Data
+│   └── ImageSets
+├── imagenet_object_localization_patched2019.tar.gz
+├── LOC_sample_submission.csv
+├── LOC_synset_mapping.txt
+├── LOC_train_solution.csv
+└── LOC_val_solution.csv
+```
+
+### Evaluation
+
+Run evaluation with the following command:
+
+```shell
+python eval.py -m mobilenet -d imagenet -dr /path/to/imagenet
+```
+
diff --git a/tools/eval/datasets/__init__.py b/tools/eval/datasets/__init__.py
@@ -0,0 +1,15 @@
+from .imagenet import ImageNet
+
+class Registery:
+    def __init__(self, name):
+        self._name = name
+        self._dict = dict()
+
+    def get(self, key):
+        return self._dict[key]
+
+    def register(self, item):
+        self._dict[item.__name__] = item
+
+DATASETS = Registery("Datasets")
+DATASETS.register(ImageNet)
diff --git a/tools/eval/datasets/imagenet.py b/tools/eval/datasets/imagenet.py
@@ -0,0 +1,64 @@
+import os
+
+import numpy as np
+import cv2 as cv
+
+from tqdm import tqdm
+
+class ImageNet:
+    def __init__(self, root, size=224):
+        self.root = root
+        self.size = size
+        self.top1_acc = -1
+        self.top5_acc = -1
+
+        self.root_val = os.path.join(self.root, "ILSVRC", "Data", "CLS-LOC", "val")
+        self.val_label_file = os.path.join(self.root, "caffe_ilsvrc12", "val.txt")
+
+        self.val_label = self.load_label(self.val_label_file)
+
+    @property
+    def name(self):
+        return self.__class__.__name__
+
+    def load_label(self, label_file):
+        label = list()
+        with open(label_file, "r") as f:
+            for line in f:
+                line = line.strip()
+                key, value = line.split()
+
+                key = os.path.join(self.root_val, key)
+                value = int(value)
+
+                label.append([key, value])
+
+        return label
+
+    def eval(self, model):
+        top_1_hits = 0
+        top_5_hits = 0
+        pbar = tqdm(self.val_label)
+        for fn, label in pbar:
+            pbar.set_description("Evaluating {} with {} val set".format(model.name, self.name))
+
+            img = cv.imread(fn)
+            img = cv.cvtColor(img, cv.COLOR_BGR2RGB)
+            img = cv.resize(img, dsize=(256, 256))
+            img = img[16:240, 16:240, :]
+
+            pred = model.infer(img)
+            if label == pred[0][0]:
+                top_1_hits += 1
+            if label in pred[0]:
+                top_5_hits += 1
+
+        self.top1_acc = top_1_hits/(len(self.val_label) * 1.0)
+        self.top5_acc = top_5_hits/(len(self.val_label) * 1.0)
+
+    def get_result(self):
+        return self.top1_acc, self.top5_acc
+
+    def print_result(self):
+        print("Top-1 Accuracy: {:.2f}%; Top-5 Accuracy: {:.2f}%".format(self.top1_acc*100, self.top5_acc*100))
+
diff --git a/tools/eval/eval.py b/tools/eval/eval.py
