makefile for deployment to ECR

stanfordnmbl · Jun 18, 2024 · d4384e9 · d4384e9
1 parent 04f8079
commit d4384e9
Show file tree

Hide file tree

Showing 6 changed files with 55 additions and 211 deletions.
diff --git a/.github/workflows/README.md b/.github/workflows/README.md
@@ -0,0 +1,4 @@
+# Workflows
+
+The actions only push the opencap (-dev) image to ECR, not the openpose (-dev) and mmpose (-dev) images.
+The mmpose (-dev) image is too big and the action fail. To push them to ECR, do it manually through the makefile by running make build and then make run.
diff --git a/.github/workflows/ecr-mmpose-dev.yml b/.github/workflows/ecr-mmpose-dev.yml
diff --git a/.github/workflows/ecr-openpose-dev.yml b/.github/workflows/ecr-openpose-dev.yml
diff --git a/docker/Makefile b/docker/Makefile
@@ -1,17 +1,38 @@
+BASE_NAME := 660440363484.dkr.ecr.us-west-2.amazonaws.com
+REPO_NAME := opencap
+PROD_BRANCH := main
+
+# Determine the branch name
+CURRENT_BRANCH := $(shell git rev-parse --abbrev-ref HEAD)
+
+# Determine image tag based on branch
+ifeq ($(CURRENT_BRANCH),$(PROD_BRANCH))
+    OPENCAP_IMAGE_TAG := opencap
+	OPENPOSE_IMAGE_TAG := openpose
+	MMPOSE_IMAGE_TAG := mmpose
+else
+    OPENCAP_IMAGE_TAG := opencap-dev
+	OPENPOSE_IMAGE_TAG := openpose-dev
+	MMPOSE_IMAGE_TAG := mmpose-dev
+endif
+
+
 .PHONY: build
 build:
 	wget -c -O ../mmpose/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth https://mc-opencap-public.s3.us-west-2.amazonaws.com/mmpose_pth/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth
 	wget -c -O ../mmpose/hrnet_w48_coco_wholebody_384x288_dark-f5726563_20200918.pth https://mc-opencap-public.s3.us-west-2.amazonaws.com/mmpose_pth/hrnet_w48_coco_wholebody_384x288_dark-f5726563_20200918.pth
-	docker build -t 660440363484.dkr.ecr.us-west-2.amazonaws.com/opencap/opencap .. -f Dockerfile
-	docker build -t 660440363484.dkr.ecr.us-west-2.amazonaws.com/opencap/openpose .. -f openpose/Dockerfile
-	docker build -t 660440363484.dkr.ecr.us-west-2.amazonaws.com/opencap/mmpose .. -f mmpose/Dockerfile
+
+	docker build -t $(BASE_NAME)/$(REPO_NAME)/$(OPENCAP_IMAGE_TAG) .. -f Dockerfile
+	docker build -t $(BASE_NAME)/$(REPO_NAME)/$(OPENPOSE_IMAGE_TAG) .. -f openpose/Dockerfile
+	docker build -t $(BASE_NAME)/$(REPO_NAME)/$(MMPOSE_IMAGE_TAG) .. -f mmpose/Dockerfile
 
 .PHONY: push
 push:
 	aws ecr get-login-password --region us-west-2 --profile opencap | docker login --username AWS --password-stdin 660440363484.dkr.ecr.us-west-2.amazonaws.com
-	docker push 660440363484.dkr.ecr.us-west-2.amazonaws.com/opencap/opencap
-	docker push 660440363484.dkr.ecr.us-west-2.amazonaws.com/opencap/openpose
-	docker push 660440363484.dkr.ecr.us-west-2.amazonaws.com/opencap/mmpose
+
+	docker push $(BASE_NAME)/$(REPO_NAME)/$(OPENCAP_IMAGE_TAG)
+	docker push $(BASE_NAME)/$(REPO_NAME)/$(OPENPOSE_IMAGE_TAG)
+	docker push $(BASE_NAME)/$(REPO_NAME)/$(MMPOSE_IMAGE_TAG)
 
 .PHONY: run
 run:

diff --git a/mmpose/loop_mmpose.py b/mmpose/loop_mmpose.py
@@ -9,45 +9,33 @@
 
 logging.basicConfig(level=logging.INFO)
 
-logging.info("Waiting TEST for data...")
+logging.info("Waiting for data...")
 
-# def checkCudaPyTorch():
-#     if torch.cuda.is_available():
-#         num_gpus = torch.cuda.device_count()
-#         logging.info(f"Found {num_gpus} GPU(s).")
-#     else:
-#         logging.info("No GPU detected. Exiting.")
-#         raise Exception("No GPU detected. Exiting.")
+def checkCudaPyTorch():
+    if torch.cuda.is_available():
+        num_gpus = torch.cuda.device_count()
+        logging.info(f"Found {num_gpus} GPU(s).")
+    else:
+        logging.info("No GPU detected. Exiting.")
+        raise Exception("No GPU detected. Exiting.")
 
 video_path = "/mmpose/data/video_mmpose.mov"
 output_dir = "/mmpose/data/output_mmpose"
 
 generateVideo=False
 
-logging.info("Before open")
 with open('/mmpose/defaultOpenCapSettings.json') as f:
     defaultOpenCapSettings = json.load(f)
 bbox_thr = defaultOpenCapSettings['hrnet']
 model_config_person='/mmpose/faster_rcnn_r50_fpn_coco.py'
 model_ckpt_person='/mmpose/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth'
 model_config_pose='/mmpose/hrnet_w48_coco_wholebody_384x288_dark_plus.py'
 model_ckpt_pose='/mmpose/hrnet_w48_coco_wholebody_384x288_dark-f5726563_20200918.pth'
-
-logging.info("isfile")
+
 if os.path.isfile(video_path):
     os.remove(video_path)
 
-# checkCudaPyTorch()
-logging.info("Before cuda check")
-if torch.cuda.is_available():
-    logging.info("cuda check: if")
-    num_gpus = torch.cuda.device_count()
-    logging.info(f"Found {num_gpus} GPU(s).")
-else:
-    logging.info("cuda check: else")
-    logging.info("No GPU detected. Exiting.")
-    raise Exception("No GPU detected. Exiting.")
-logging.info("After cuda check")
+checkCudaPyTorch()
 while True:    
     if not os.path.isfile(video_path):
         time.sleep(0.1)
@@ -60,13 +48,7 @@
     os.makedirs(output_dir)
 
     try:
-        # checkCudaPyTorch()
-        if torch.cuda.is_available():
-            num_gpus = torch.cuda.device_count()
-            logging.info(f"Found {num_gpus} GPU(s).")
-        else:
-            logging.info("No GPU detected. Exiting.")
-            raise Exception("No GPU detected. Exiting.")
+        checkCudaPyTorch()
         # Run human detection.
         pathModelCkptPerson = model_ckpt_person
         bboxPath = os.path.join(output_dir, 'box.pkl')

diff --git a/openpose/loop_openpose.py b/openpose/loop_openpose.py
@@ -9,16 +9,16 @@
 logging.basicConfig(level=logging.INFO)
 
 #%%
-# def check_cuda_device():
-#     try:
-#         # Run the nvidia-smi command and capture the output
-#         _ = subprocess.check_output(["nvidia-smi"])
-#         # If the command ran successfully, assume a CUDA device is present
-#         logging.info("A CUDA-capable device is detected.")
-#     except subprocess.CalledProcessError as e:
-#         # If the command fails, it means no CUDA device is detected
-#         logging.info("No CUDA-capable device is detected. Error: %s", e)
-#         raise Exception("No CUDA-capable device is detected.")
+def check_cuda_device():
+    try:
+        # Run the nvidia-smi command and capture the output
+        _ = subprocess.check_output(["nvidia-smi"])
+        # If the command ran successfully, assume a CUDA device is present
+        logging.info("A CUDA-capable device is detected.")
+    except subprocess.CalledProcessError as e:
+        # If the command fails, it means no CUDA device is detected
+        logging.info("No CUDA-capable device is detected. Error:", e)
+        raise Exception("No CUDA-capable device is detected.")
 
 #%%
 def getVideoOrientation(videoPath):
@@ -74,21 +74,7 @@ def getResolutionCommand(resolutionPoseDetection, horizontal):
 
 #%% 
 
-logging.info("Waiting TEST for data...")
-# check_cuda_device()
-logging.info("Before cuda check")
-try:
-    logging.info("cuda check: try")
-    # Run the nvidia-smi command and capture the output
-    _ = subprocess.check_output(["nvidia-smi"])
-    # If the command ran successfully, assume a CUDA device is present
-    logging.info("A CUDA-capable device is detected.")
-except subprocess.CalledProcessError as e:
-    logging.info("cuda check: except")
-    # If the command fails, it means no CUDA device is detected
-    logging.info("No CUDA-capable device is detected. Error: %s", e)
-    raise Exception("No CUDA-capable device is detected.")
-logging.info("After cuda check")
+logging.info("Waiting for data...")
 
 video_path = "/openpose/data/video_openpose.mov"
 output_dir = "/openpose/data/output_openpose"
@@ -106,7 +92,7 @@ def getResolutionCommand(resolutionPoseDetection, horizontal):
         time.sleep(0.1)
         continue
 
-    logging.info("Processing openpose ...")
+    logging.info("Processing...")
 
     if os.path.isdir(output_dir):
         shutil.rmtree(output_dir)
@@ -115,16 +101,7 @@ def getResolutionCommand(resolutionPoseDetection, horizontal):
     horizontal = getVideoOrientation(video_path)
     cmd_hr = getResolutionCommand(resolutionPoseDetection, horizontal)
 
-    # check_cuda_device()
-    try:
-        # Run the nvidia-smi command and capture the output
-        _ = subprocess.check_output(["nvidia-smi"])
-        # If the command ran successfully, assume a CUDA device is present
-        logging.info("A CUDA-capable device is detected.")
-    except subprocess.CalledProcessError as e:
-        # If the command fails, it means no CUDA device is detected
-        logging.info("No CUDA-capable device is detected. Error:", e)
-        raise Exception("No CUDA-capable device is detected.")
+    check_cuda_device()
     command = "/openpose/build/examples/openpose/openpose.bin\
         --video {video_path}\
         --display 0\