evaluate_model_performance.py

import os

from transformers import DeformableDetrForObjectDetection

import helpers.const
from coco_detection import CocoDetection
from helpers.evaluation_functions import evaluate_result
from train_model import get_dataloader
from models import DeformableDetr

if __name__ == "__main__":
    import torch.multiprocessing

    tmpdir = os.environ.get("TMPDIR", "") # Use the TMPDIR environment variable if set, otherwise use default
    torch.multiprocessing.set_start_method("spawn", force=True)

    lm = DeformableDetr.DeformableDetr.load_from_checkpoint(checkpoint_path='prod/after_training_augtest.ckpt',
                                                            lr=1e-4, lr_backbone=1e-5, weight_decay=1e-4, id2label=helpers.const.id2label, assign=True)
    lm.model.save_pretrained('finetuned_detr')
    lm.model.config.save_pretrained('finetuned_detr')

    model = DeformableDetrForObjectDetection.from_pretrained('finetuned_detr')
    model.eval()

    # Get Dataset
    small_dataset = True
    train_dataset_name = "coco_train_small.json" if small_dataset else "coco_train.json"
    val_dataset_name = "coco_val_small.json" if small_dataset else "coco_val.json"
    train_dataloader, _, image_processor = get_dataloader(caching=False, model=model,
                                                          small_dataset=small_dataset,dataset_root=tmpdir, batch_size=helpers.const.batch_size)


#os.path.join(dataset_root, "training", "extracted_frames", "train")
    val_dataset = CocoDetection(img_folder=os.path.join(tmpdir, "training", "extracted_frames", "val"), processor=image_processor,
                                coco_json_filename=val_dataset_name, train=False)
    train_dataset = CocoDetection(img_folder=os.path.join(tmpdir, "training", "extracted_frames", "train"), processor=image_processor, coco_json_filename=train_dataset_name)

    evaluate_result(model, val_dataset, image_processor)