added flask changes to allow webserver to start

detr_object_detection/detr_hpc/Dockerfile

@@ -16,7 +16,9 @@ ENV ANNOTATION_FILE_NAME _annotations.coco.json
 ENV LOG_EVERY_N_STEP 5
 ENV MAX_NU_METRIC_IMAGES 10
 ENV DETECTION_THRESHOLD 0.8
-ENV PORT 8061
+
+ENV GRPCPORT 8061
+ENV HTTPPORT 8062
 
 
 WORKDIR /home/objectdetection

detr_object_detection/detr_hpc/app.py

@@ -10,9 +10,12 @@ import json
 import copy
 
 from logger import Logger
-import config
+
 logger = Logger(__name__)
 
+config = None
+port = int(os.getenv("HTTPPORT", "8062"))
+
 app = Flask(__name__)
 
 default_parameters_dict = {
@@ -124,9 +127,11 @@ def training_input():
     return render_template("index.html", training_var_form=training_vals_form, trainingStatusForm=trainingStatusForm, current_para=current_para)
 
 
-def app_run():
+def app_run(config_main):
+    global config
+    config = config_main
     app.secret_key = "detr"
     bootstrap = Bootstrap(app)
-    app.run(host="0.0.0.0", port=8062)
+    app.run(host="0.0.0.0", port=port)
 
 #app_run()
\ No newline at end of file

detr_object_detection/detr_hpc/config.py

@@ -25,7 +25,7 @@ class Config:
             self.log_every_n_steps = 5
             self.annotation_file_name = "_annotations.coco.json"
             self.lightning_logs_dir = "./dev/tensorboard_logs/"
-            self.dataset_path = "./dev/dataset/maritimair"
+            self.dataset_path = "./dev/dataset/maritimsmall"
             self.input_path = "./dev/inputimage"
             self.output_path = "./dev/outputimage"
             self.status_folder = "./dev/status"

detr_object_detection/detr_hpc/detr_client_test.py

@@ -34,13 +34,34 @@ def run_inference(input_path, output_path):
     
     print(response)
 
+def start_train():
+    logger.debug("Calling HPP_Stub for training")
+    addr = 'localhost:{}'.format(server_port)
+    logger.debug("connecting to {}".format(addr))
+    with grpc.insecure_channel(addr) as channel:
+        stub = model_pb2_grpc.DetrModelServicerStub(channel)
+        
+        #inference
+        ui_request = model_pb2.TrainingConfig()
+        ui_request.config = "test"
+        response = stub.startTraining(ui_request)
+
+        # training
+        # ui_request = model_pb2.TrainingConfig(config="test")
+        # response = stub.startTraining(ui_request)
+
+        logger.debug("Training finished")
+    
+    print(response)
+
 
 import argparse
 if __name__ == '__main__':
-    parser = argparse.ArgumentParser(description='Process some integers.')
-    parser.add_argument('-i','--input', type=str , help='Input Image',required=True) # will be accesible with args.OPT
-    parser.add_argument('-o','--output', type=str , help='Output Image',required=True) # will be accesible with args.OPT
+    # parser = argparse.ArgumentParser(description='Process some integers.')
+    # parser.add_argument('-i','--input', type=str , help='Input Image',required=True) # will be accesible with args.OPT
+    # parser.add_argument('-o','--output', type=str , help='Output Image',required=True) # will be accesible with args.OPT
 
-    args = parser.parse_args()
+    # args = parser.parse_args()
 
-    run_inference(args.input, args.output)
+    # run_inference(args.input, args.output)
+    start_train()

detr_object_detection/detr_hpc/detr_server.py

@@ -5,14 +5,17 @@ from concurrent import futures
 import json
 import datetime
 import os
+import threading
 
 from od.detrOD import DetrOD
 from config import Config
+from app import app_run, get_train_parameters
 
 from logger import Logger
 logger = Logger(__name__)
 
-port = int(os.getenv("PORT", "8061"))
+port = int(os.getenv("GRPCPORT", "8061"))
+SHARED_FOLDER = os.getenv("SHARED_FOLDER_PATH", "./dev2")  
 
 
 class DetrModelServicer(model_pb2_grpc.DetrModelServicer):
@@ -32,23 +35,69 @@ class DetrModelServicer(model_pb2_grpc.DetrModelServicer):
         try:
             logger.info('MetricsAvailable')
             response = model_pb2.TrainStatus()
-
-            logger.info("Starting the training...")
-            self.detrOd.load_custom_model()
-
-            self.start_count = 1
-            with open(self.config.status_folder+"/results.txt", mode="w") as f:
-                f.write("Training started. Please wait until its finished")
-                f.close()
-            self.detrOd.train()
-            logger.info("Training finished...")
-            with open(self.config.status_folder+"/results.txt", mode="w") as f:
-                f.write("Training finished.")
-                f.close()
-            self.start_count = 2
-            response.status = self.start_count
-            self.get_metrics_metadata(None,None)
-            return response
+            # states:
+            # 0: training not started.
+            # 1: being trained.
+            # 2: training done.
+            # -1: Error
+            if self.start_count == 0:
+                logger.info("Starting the training...")
+                self.detrOd.load_custom_model()
+                parameters = get_train_parameters()
+                logger.info("Got parameters")
+                logger.info(parameters)
+                self.start_count = 1
+                with open(self.config.status_folder+"/results.txt", mode="w") as f:
+                    f.write("Training started. Please wait until its finished")
+                    f.close()
+                self.detrOd.train(
+                    lr=parameters["lr"], 
+                    lr_backbone=parameters["lr_backbone"], 
+                    weight_decay=parameters["weight_decay"],
+                    max_epoch=parameters["max_epoch"],
+                    gradient_clip_val=parameters["gradient_clip_val"],
+                    accumulate_grad_batches=parameters["accumulate_grad_batches"],
+                    batch_size = parameters["batch_size"],
+                    dataloader_workers = parameters["dataloader_workers"]
+                )
+                logger.info("Training finished...")
+                with open(self.config.status_folder+"/results.txt", mode="w") as f:
+                    f.write("Training finished.")
+                    f.close()
+                self.start_count = 2
+                response.status = self.start_count
+                self.get_metrics_metadata(None,None)
+                return response
+
+            elif self.start_count == 1:
+                context.set_code(grpc.StatusCode.NOT_FOUND)
+                message = "Training is started and not finished yet."
+                context.set_details(message)
+                with open(self.config.status_folder+"/results.txt", mode="w") as f:
+                        f.write(message)
+                        f.close()
+                response.status = self.start_count
+                return response
+
+            elif self.start_count == 2:
+                message = "Training finished. Download your model from sharedfolder. Please return to the home page and start the training with the new parameters."
+                logger.info(message)
+                with open(self.config.status_folder+"/results.txt", mode="w") as f:
+                    f.write(message)
+                    f.close()
+                self.start_count = 0
+                response.status = self.start_count
+                return response
+            
+            else:
+                message = "Something went wrong. Please check the container or contact the publisher."
+                logger.info(message)
+                with open(self.config.status_folder+"/results.txt", mode="w") as f:
+                    f.write(message)
+                    f.close()
+                self.start_count = -1
+                response.status = self.start_count
+                return response
         except Exception as e:
             logger.error("Error while running training")
             logger.error(e, exc_info=True)
@@ -111,17 +160,18 @@ class DetrModelServicer(model_pb2_grpc.DetrModelServicer):
             context.set_code(grpc.StatusCode.NOT_FOUND)
 
 
-def serve(config):
+def serve_grpc(config):
     server = grpc.server(futures.ThreadPoolExecutor(max_workers=10))
     model_pb2_grpc.add_DetrModelServicerServicer_to_server(DetrModelServicer(config), server)
     server.add_insecure_port('[::]:{}'.format(port))
     server.start()
     logger.info("Start config web-ui")
-    server.wait_for_termination()
+    return server
 
 
 if __name__ == '__main__':
-    folder_path = "./dev2"
     prod_flag = os.getenv("PRODUCTION", False)
-    config = Config(prod_flag,folder_path)
-    serve(config)
+    config = Config(prod_flag, SHARED_FOLDER)
+    grpc_server = serve_grpc(config)
\ No newline at end of file
+    threading.Thread(target=app_run(config)).start()
+    grpc_server.wait_for_termination()
\ No newline at end of file