evaluate_ecsam.py

import numpy as np
from edge_device import EdgeDevice
from cloud_server import CloudServer

def compute_iou(pred_mask, gt_mask):
    intersection = np.logical_and(pred_mask, gt_mask).sum()
    union = np.logical_or(pred_mask, gt_mask).sum()
    if union == 0:
        return 1.0 if intersection == 0 else 0.0
    return intersection / union

def compute_pixel_accuracy(pred_mask, gt_mask):
    return (pred_mask == gt_mask).sum() / pred_mask.size

def main():
    # Initialize ECSAM system
    edge = EdgeDevice()
    cloud = CloudServer()

    # Run ECSAM pipeline (simulate one run)
    frames, prompts = edge.input_handler.receive_input()
    visual_emb = edge.visual_prompt_encoder.encode(prompts)
    split_point = edge.workload_partitioner.partition(model_layers=1000, resources=edge.resources, latency_constraint=edge.resources.max_latency_ms)
    local_feat, cloud_state = edge.image_encoder.encode(frames, prompts, split_point)
    edge.communicator.send_to_cloud(cloud_state)
    cloud_emb = cloud.process_from_edge(cloud_state)
    pred_mask = edge.embedding_decoder.decode(visual_emb, cloud_emb)

    # Simulate a ground truth mask (random binary mask for demonstration)
    gt_mask = np.random.randint(0, 2, size=pred_mask.shape, dtype=np.uint8)

    # Compute evaluation metrics
    iou = compute_iou(pred_mask, gt_mask)
    acc = compute_pixel_accuracy(pred_mask, gt_mask)

    print(f"[Eval] IoU: {iou:.4f}")
    print(f"[Eval] Pixel Accuracy: {acc:.4f}")

if __name__ == "__main__":
    main()