testing dynamic onnx

app.py

@@ -4,7 +4,7 @@ from tasks import text, image, audio
 
 # Load environment variables
 load_dotenv()
-
+print(1)
 app = FastAPI(
     title="Frugal AI Challenge API",
     description="API for the Frugal AI Challenge evaluation endpoints"
@@ -12,8 +12,11 @@ app = FastAPI(
 
 # Include all routers
 app.include_router(text.router)
+print(2)
 app.include_router(image.router)
+print(3)
 app.include_router(audio.router)
+print(4)
 
 @app.get("/")
 async def root():

requirements.txt

@@ -14,4 +14,9 @@ ultralytics-thop==2.0.14
 #opencv-python==4.11.0.86
 python-dotenv==1.0.0
 onnxruntime==1.19.2
-matplotlib==3.8.1
+matplotlib==3.8.1
+onnx==1.17.0
+tensorrt==10.8.0.43
+tensorrt_cu12==10.8.0.43
+tensorrt_cu12_bindings==10.8.0.43
+tensorrt_cu12_libs==10.8.0.43

tasks/image.py

@@ -1,25 +1,38 @@
 from fastapi import APIRouter
+print(1)
 from datetime import datetime
+print(2)
 from datasets import load_dataset
+print(3)
 import numpy as np
+print(4)
 from sklearn.metrics import accuracy_score, precision_score, recall_score
+print(5)
 import random
+print(6)
 import os
+print(7)
 from ultralytics import YOLO
+print(8)
 
 from .utils.evaluation import ImageEvaluationRequest
+print(9)
 from .utils.emissions import tracker, clean_emissions_data, get_space_info
+print(10)
 
 from dotenv import load_dotenv
+print(11)
 load_dotenv()
+print(12)
 
 router = APIRouter()
+print(13)
 
 #MODEL_TYPE = "YOLOv11n"
 DESCRIPTION = f"YOLOv8n model with batch 128 inference on CPU"
-
+print(14)
 ROUTE = "/image"
-
+print(15)
 def parse_boxes(annotation_string):
     """Parse multiple boxes from a single annotation string.
     Each box has 5 values: class_id, x_center, y_center, width, height"""
@@ -32,7 +45,7 @@ def parse_boxes(annotation_string):
             box = values[i+1:i+5]
             boxes.append(box)
     return boxes
-
+print(16)   
 def compute_iou(box1, box2):
     """Compute Intersection over Union (IoU) between two YOLO format boxes."""
     # Convert YOLO format (x_center, y_center, width, height) to corners
@@ -61,7 +74,7 @@ def compute_iou(box1, box2):
     union = box1_area + box2_area - intersection
     
     return intersection / (union + 1e-6)
-
+print(17)
 def compute_max_iou(true_boxes, pred_box):
     """Compute maximum IoU between a predicted box and all true boxes"""
     max_iou = 0
@@ -69,17 +82,17 @@ def compute_max_iou(true_boxes, pred_box):
         iou = compute_iou(true_box, pred_box)
         max_iou = max(max_iou, iou)
     return max_iou
-
+print(18)
 def load_model(path_to_model, model_type="YOLO"):
     if model_type == "YOLO":
         model = YOLO(path_to_model)
     else:
         raise NotImplementedError
     return model
-
+print(19)
 def get_boxes_list(predictions):
     return [box.tolist() for box in predictions.boxes.xywhn]
-
+print(20)
 @router.post(ROUTE, tags=["Image Task"],
              description=DESCRIPTION)
 async def evaluate_image(request: ImageEvaluationRequest):
@@ -120,11 +133,11 @@ async def evaluate_image(request: ImageEvaluationRequest):
     #PATH_TO_MODEL = 'models/best_YOLOv11n_1280.onnx'
     #PATH_TO_MODEL = 'models/best_yolov6n_1280.pt'
     #PATH_TO_MODEL = 'models/best_YOLOv11n_1280_real_half.onnx'
-    PATH_TO_MODEL = 'models/best_YOLOv8n.onnx'
+    PATH_TO_MODEL = 'models/best_YOLOv8n_dynamic.onnx'
     INFERENCE_ENGINE_TYPE = 'pt'
     INPUT_SIZE = 640
     N_TEST_BATCHES = 2
-    BATCH_SIZE = 128 # Can be adjusted as needed
+    BATCH_SIZE = 64 # Can be adjusted as needed
 
     def preprocessor(frame):
         #frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) # Only when read from file
@@ -136,8 +149,8 @@ async def evaluate_image(request: ImageEvaluationRequest):
 
     class Inference:
         def __init__(self, model, image):
-            self.session = onnxruntime.InferenceSession(model,#, providers=["CPUExecutionProvider"]
-                                                        providers=["CUDAExecutionProvider"]
+            self.session = onnxruntime.InferenceSession(model, providers=["CPUExecutionProvider"]
+                                                        #providers=["CUDAExecutionProvider"]
                                                         )
             model_inputs = self.session.get_inputs()
             input_shape = model_inputs[0].shape