43 lines
1.4 KiB
Python
43 lines
1.4 KiB
Python
from paddle.inference import create_predictor
|
|
from paddle.inference import Config
|
|
from ppdet.infer import Detector
|
|
from ppdet.visualize import visualize_box_mask
|
|
import os, yaml, logging
|
|
from huggingface_hub import snapshot_download
|
|
|
|
class PPDet:
|
|
def __init__(self):
|
|
model_dir = snapshot_download(repo_id="InfiniFlow/picodet_lcnet_x1_0_fgd_layout_cdla")
|
|
self.mdl = Detector(model_dir)
|
|
with open(os.path.join(model_dir, 'infer_cfg.yml'), "r") as f:
|
|
self.conf = yaml.safe_load(f)
|
|
|
|
def __call__(self, image_list:list, thr=0.7):
|
|
result = self.mdl.predict_image(image_list, visual=False)
|
|
cate = self.conf['label_list']
|
|
start_idx = 0
|
|
res = []
|
|
for idx, img in enumerate(image_list):
|
|
im_bboxes_num = result['boxes_num'][idx]
|
|
bb = []
|
|
for b in result["boxes"][start_idx:start_idx +im_bboxes_num]:
|
|
clsid, bbox, score = int(b[0]), b[2:], b[1]
|
|
if score < thr:continue
|
|
if clsid >= len(cate):
|
|
logging.warning(f"bad category id")
|
|
continue
|
|
bb.append({
|
|
"type": cate[clsid].lower(),
|
|
"bbox": bbox,
|
|
"score": score
|
|
})
|
|
|
|
res.append(bb)
|
|
start_idx += im_bboxes_num
|
|
|
|
return res
|
|
|
|
|
|
|
|
|
|
|