inference.py

from mmdet.evaluation.metrics.coco_metric import CocoMetric
import os
from mmdet.apis import DetInferencer

root_dir = r'./experiments'
batch_size = 8
experiments_to_infer = [
    f"FD_diff", #

    f"FD_load_deform", #
    f"FD_load_dino", #
    f"FD_load_diff", #

    f"FD_glip", #
    f"FD_gdino", #

    f"FD_BCG_glip",  #
    f"FD_BCG_gdino", #
    f"FD_BCG_SOS", #

]

class_name_1 = (
    "posterior teeth",
    "anterior teeth",
    "anterior teeth No FD",
    "anterior teeth FD",
)

coco_1 = CocoMetric(
    ann_file=r'data/v1/40_FD_BCG_test.json'
    , classwise=True)

coco_1.cat_ids = coco_1._coco_api.get_cat_ids(cat_names=list(class_name_1))

coco_1.img_ids = coco_1._coco_api.get_img_ids()

test_image_ids = coco_1.img_ids
coco_images = []
coco_annotations = []
image_id, annotation_id = 1, 1
all_list = []
for img_idx in test_image_ids:
    image_meta_info = coco_1._coco_api.loadImgs(img_idx)
    image_file = image_meta_info[0]['file_name']
    all_list.append(os.path.join('data/v1/images_all', image_file))

for _work_dir in experiments_to_infer:
    work_dir = os.path.join(root_dir, _work_dir)
    if os.path.exists(work_dir):
        out_work_dir = work_dir.split('/')[-1]

        best_checkpoint = [i for i in os.listdir(work_dir) if i.startswith('best_')][-1]
        config_checkpoint = [i for i in os.listdir(work_dir) if i.endswith('.py')][0]
        # %%
        config_path = os.path.join(work_dir, config_checkpoint)

        # Setup a checkpoint file to load
        checkpoint = os.path.join(work_dir, best_checkpoint)

        # Initialize the DetInferencer
        inferencer = DetInferencer(model=config_path, weights=checkpoint, device='cuda:0')

        if 'BCG' in out_work_dir:
            texts = "posterior teeth. anterior teeth. anterior teeth No FD. anterior teeth FD"
        else:
            texts = "anterior teeth No FD. anterior teeth FD"

        _ = inferencer(all_list,
                       texts=texts,
                       custom_entities=True,
                       out_dir=f"predictions/{out_work_dir}/", no_save_pred=False, batch_size=batch_size,
                       no_save_vis=True)
    else:
        print(f'Path {work_dir} is not existing')