added roboflow example

examples/roboflow-logging/client-multiple-endpoints.py

@@ -0,0 +1,42 @@
+import cv2, requests, json, argparse
+import matplotlib.pyplot as plt
+
+parser = argparse.ArgumentParser(description="Pass a name to a file to be annotated")
+parser.add_argument('--image_path', default="test/images/a9f16c_2_9_png.rf.c048a60764e56735d7465cdec974d102.jpg")
+
+MODEL_IMG_SIZE = 640
+ENDPOINT_URL_FINETUNED = "http://localhost:5543/yolov5-s-finetuned/predict/from_files"
+ENDPOINT_URL_COCO = "http://localhost:5543/yolov5-s-coco/predict/from_files"
+
+def main(image_path):
+    for endpoint_url in [ENDPOINT_URL_FINETUNED, ENDPOINT_URL_COCO]:
+        im = cv2.imread(image_path)
+        im_size = im.shape[:2]
+        assert im_size[0] == im_size[1]
+        scale_ratio = im_size[0] / MODEL_IMG_SIZE
+
+        resp = requests.post(
+            url=endpoint_url,
+            files=[('request', open(image_path, 'rb'))]
+        )
+
+        boxes = json.loads(resp.text)['boxes'][0]
+        for xmin, ymin, xmax, ymax in boxes:
+            start_point = (int(xmin * scale_ratio), int(ymin * scale_ratio))
+            end_point = (int(xmax * scale_ratio), int(ymax * scale_ratio))
+            color = (0, 255, 0)
+            thickness = 2
+            im = cv2.rectangle(im, start_point, end_point, color, thickness)
+
+        plt.figure(figsize=(15, 15))
+        plt.axis("off")
+        plt.imshow(im)
+
+        if endpoint_url == ENDPOINT_URL_FINETUNED:
+            plt.savefig("annotated-finetuned.png")
+        else:
+            plt.savefig("annotated-coco.png")
+
+if __name__ == '__main__':
+    args = parser.parse_args()
+    main(args.image_path)

examples/roboflow-logging/client-send-all.py

@@ -0,0 +1,27 @@
+import requests, os, time
+
+# the dataset we downloaded had 3 subsets in 3 folders
+paths = []
+for folder_name in ['test', 'valid', 'train']:
+    path = f"{folder_name}/images/"
+    paths += [path + img_name for img_name in os.listdir(path)]
+
+
+# same URL for the endpoint as before
+ENDPOINT_URL = "http://localhost:5543/yolov5-s-coco/predict/from_files"
+
+
+# send each image to the endpoint
+i = 0
+print(f"Sending {len(paths)} images to the server")
+for image_path in paths:
+    if i % 60 == 0:
+        print(i)
+    i+=1
+
+    resp = requests.post(
+      url=ENDPOINT_URL,
+      files=[('request', open(image_path, 'rb'))]
+    )
+    print(resp)
+    time.sleep(1)

examples/roboflow-logging/client-send-one.py

@@ -0,0 +1,11 @@
+import requests, json
+
+ENDPOINT_URL = "http://localhost:5543/yolov5-s-coco/predict/from_files"
+IMAGE_PATH = "test/images/4b770a_3_6_png.rf.f5d975605c1f73e1a95a1d8edc4ce5b1.jpg"
+
+resp = requests.post(
+  url=ENDPOINT_URL,
+  files=[('request', open(IMAGE_PATH, 'rb'))]
+)
+
+print(json.loads(resp.text))

examples/roboflow-logging/server/requirements.txt

@@ -0,0 +1,2 @@
+deepsparse[server,yolo]
+roboflow

examples/roboflow-logging/server/roboflow-logger.py

@@ -0,0 +1,28 @@
+from deepsparse.loggers import BaseLogger, MetricCategories
+import typing, PIL, io, requests, datetime
+from requests_toolbelt.multipart.encoder import MultipartEncoder
+
+class RoboflowLogger(BaseLogger):
+
+    # the arguments to the construction will be defined in the config file
+    def __init__(self, dataset_name: str, api_key: str):
+        # per Roboflow docs
+        self.upload_url = f"https://api.roboflow.com/dataset/{dataset_name}/upload?api_key={api_key}"
+        super(RoboflowLogger, self).__init__()
+
+    # this function will be called from DeepSparse Server, based on the config
+    def log(self, identifier: str, value: typing.Any, category: typing.Optional[str]=None):
+        if category == MetricCategories.DATA:
+            # unpacks value and converts to image in a buffer          
+            img = PIL.Image.fromarray(value.images[0], mode="RGB")
+            buffered = io.BytesIO()
+            img.save(buffered, quality=90, format="JPEG")
+
+            # packs as multipart
+            img_name = f"production-image-{datetime.datetime.now()}.jpg"
+            m =  MultipartEncoder(fields={'file': (img_name, buffered.getvalue(), "image/jpeg")})
+
+            # uploads to roboflow
+            r = requests.post(self.upload_url, data=m, headers={'Content-Type': m.content_type})
+
+            print("request_complete")

examples/roboflow-logging/server/server-config-no-roboflow-logging.yaml

@@ -0,0 +1,9 @@
+## server-config-no-roboflow-logging.yaml
+loggers: 
+    python:
+
+endpoints:
+  - task: yolo
+    model: zoo:cv/detection/yolov5-s/pytorch/ultralytics/coco/pruned_quant-aggressive_94
+    name: yolov5-s-coco
+    route: /yolov5-s-coco/predict

examples/roboflow-logging/server/server-config-roboflow-logging.yaml

@@ -0,0 +1,18 @@
+# server-config-roboflow-logging.yaml
+loggers:
+  roboflow_logger:
+    path: server/roboflow-logger.py:RoboflowLogger
+    api_key: YOUR_API_KEY
+    dataset_name: YOUR_DATASET_NAME
+
+endpoints:
+  - task: yolo
+    model: zoo:cv/detection/yolov5-s/pytorch/ultralytics/coco/pruned_quant-aggressive_94
+    name: yolov5-s-coco
+    route: /yolov5-s-coco/predict
+    data_logging:
+      pipeline_inputs:
+        - func: identity
+          frequency: 1
+          target_loggers:
+            - roboflow_logger