执行测试文件报错，keyerror：‘img_name’

[lyx-0213]

请问怎么解决啊谢谢

[xingchen-yao]

遇到了同样的问题，请问解决了吗

[wuhengliangliang]

请问问题解决了嘛

[tiamjiakun]

在pan_ic15.py中的‘prepare_test_data’方法（第441行之后）中添加 img_name = img_path.split("/")[-1] img_meta.update(dict(img_path=img_path)) img_meta.update(dict(img_name=img_name)) 这三行代码，然后就好了。

[zeng-cy]

prepare_test_data

您好，能具体说一下这个文件在哪里吗？没找到

[894900372]

I also encountered this issue during the testing of ctw1500, and the reason for the problem is that there is indeed no 'imgname' key in the img_metas dictionary. This may be because the image name was not correctly set or passed during the data loading or processing phase. I have resolved this issue and need to modify the content of utils/result_format.py: class ResultFormat(object): def init(self, data_type, result_path): self.data_type = data_type self.result_path = result_path Self. img_index=0 # Initialize image index

if osp.isfile(result_path): os.remove(result_path)

if result_path.endswith('.zip'): result_path = result_path.replace('.zip', '')

if not osp.exists(result_path): os.makedirs(result_path)

def write_result(self, imgmetas, outputs): Imgname=f'image {self. img_index} '# Generate image names using indexes Self. img_index+=1 # Update index

if 'IC15' in self.data_type: self._write_result_ic15(img_name, outputs) elif 'TT' in self.data_type: self._write_result_tt(img_name, outputs) elif 'CTW' in self.data_type: self._write_result_ctw(img_name, outputs) elif 'MSRA' in self.data_type: self._write_result_msra(img_name, outputs) If you want to debug to see if this key is included, you can add the following code in test.py: def test(test_loader, model, cfg): model.eval()

with_rec = hasattr(cfg.model, 'recognition_head') if with_rec: pp = Corrector(cfg.data.test.type, **cfg.test_cfg.rec_post_process)

if cfg.vis: vis = Visualizer(vis_path=osp.join('vis/', cfg.data.test.type))

rf = ResultFormat(cfg.data.test.type, cfg.test_cfg.result_path)

if cfg.report_speed: speed_meters = dict( backbone_time=AverageMeter(500), neck_time=AverageMeter(500), det_head_time=AverageMeter(500), det_post_time=AverageMeter(500), rec_time=AverageMeter(500), total_time=AverageMeter(500))

print('Start testing %d images' % len(test_loader)) for idx, data in enumerate(test_loader): print('Testing %d/%d\r' % (idx, len(test_loader)), end='', flush=True)

Print img_metas to see what's inside

print("img_metas:", data['img_metas'])

prepare input

data['imgs'] = data['imgs'].cuda() data.update(dict(cfg=cfg))

forward

with torch.no_grad(): outputs = model(**data)

if cfg.report_speed: report_speed(outputs, speed_meters)

post process of recognition

if with_rec: outputs = pp.process(data['img_metas'], outputs)

save result

rf.write_result(data['img_metas'], outputs)

visualize

if cfg.vis: vis.process(data['img_metas'], outputs)

print('Done!')