"list index out of range" when taking single-image input

[Claber]

@lonestar234028
code position:
https://github.com/lonestar234028/modelscope/blob/master/modelscope/preprocessors/ofa/visual_question_answering.py#L113

call stack log lines:
/root/code/nlvr/modelscope/reason_step_huge/test_ok_vqa_with_rationals_retri │
│ eval.py:157 in │
│ │
│ 154 │ │ │ │ q_and_a.update({"question_id":int(q.question_id)}) │
│ 155 │ │ │ │ input_batch.append(input_from_question(q)) │
│ 156 │ │ │ │ output_batch.append(q_and_a) │
│ ❱ 157 │ │ │ answer_batch = ofa_pipe(input_batch) │
│ 158 │ │ │ for a, p in zip(answer_batch, output_batch): │
│ 159 │ │ │ │ p.update({"answer" :a[OutputKeys.TEXT][0]}) │
│ 160 │ │ │ │ res.append(p) │
│ │
│ /opt/conda/lib/python3.7/site-packages/modelscope/pipelines/base.py:189 in │
│ call │
│ │
│ 186 │ │ │ if batch_size is None: │
│ 187 │ │ │ │ output = [] │
│ 188 │ │ │ │ for ele in input: │
│ ❱ 189 │ │ │ │ │ output.append(self._process_single(ele, *args, **k │
│ 190 │ │ │ else: │
│ 191 │ │ │ │ output = self._process_batch(input, batch_size, **kwar │
│ 192 │
│ │
│ /opt/conda/lib/python3.7/site-packages/modelscope/pipelines/base.py:228 in │
│ _process_single │
│ │
│ 225 │ │ import copy │
│ 226 │ │ │
│ 227 │ │ all_text_encoders_input = [] │
│ ❱ 228 │ │ out = self.preprocess(input, **preprocess_params) │
│ 229 │ │ if isinstance(input['text'], list): │
│ 230 │ │ │ for item in input['text']: │
│ 231 │ │ │ │ tmp_e = copy.deepcopy(input) │
│ │
│ /opt/conda/lib/python3.7/site-packages/modelscope/pipelines/base.py:365 in │
│ preprocess │
│ │
│ 362 │ │ assert self.preprocessor is not None, 'preprocess method shoul │
│ 363 │ │ assert not isinstance(self.preprocessor, List),\ │
│ 364 │ │ │ 'default implementation does not support using multiple pr │
│ ❱ 365 │ │ return self.preprocessor(inputs, **preprocess_params) │
│ 366 │ │
│ 367 │ def forward(self, inputs: Dict[str, Any], │
│ 368 │ │ │ │ **forward_params) -> Dict[str, Any]: │
│ │
│ /opt/conda/lib/python3.7/site-packages/modelscope/preprocessors/multi_modal. │
│ py:147 in call │
│ │
│ 144 │ │ │ data = input │
│ 145 │ │ else: │
│ 146 │ │ │ data = self._build_dict(input) │
│ ❱ 147 │ │ sample = self.preprocess(data) │
│ 148 │ │ str_data = dict() │
│ 149 │ │ for k, v in data.items(): │
│ 150 │ │ │ str_data[k] = str(v) │
│ │
│ /opt/conda/lib/python3.7/site-packages/modelscope/preprocessors/ofa/visual_q │
│ uestion_answering.py:47 in call │
│ │
│ 44 │ │ if self.mode == ModeKeys.TRAIN: │
│ 45 │ │ │ return self._build_train_sample(data) │
│ 46 │ │ else: │
│ ❱ 47 │ │ │ return self._build_infer_sample(data) │
│ 48 │ │
│ 49 │ def _build_train_sample(self, data: Dict[str, Any]) -> Dict[str, A │
│ 50 │ │ r""" │
│ │
│ /opt/conda/lib/python3.7/site-packages/modelscope/preprocessors/ofa/visual_q │
│ uestion_answering.py:140 in _build_infer_sample │
│ │
│ 137 │ def _build_infer_sample(self, data: Dict[str, Any]) -> Dict[str, A │
│ 138 │ │ img_info = data[self.column_map['image']] │
│ 139 │ │ if isinstance(img_info, list): │
│ ❱ 140 │ │ │ return self._build_infer_sample_list(data) │
│ 141 │ │ image = self.get_img_pil(img_info) │
│ 142 │ │ │
│ 143 │ │ patch_image = self.patch_resize_transform(image) │
│ │
│ /opt/conda/lib/python3.7/site-packages/modelscope/preprocessors/ofa/visual_q │
│ uestion_answering.py:113 in _build_infer_sample_list │
│ │
│ 110 │ │
│ 111 │ def _build_infer_sample_list(self, data: Dict[str, Any]) -> Dict[s │
│ 112 │ │ image0 = self.get_img_pil(data[self.column_map['image']][0]) │
│ ❱ 113 │ │ image1 = self.get_img_pil(data[self.column_map['image']][1]) │
│ 114 │ │ patch_image = self.patch_resize_transform(image0) │
│ 115 │ │ patch_image_2 = self.patch_resize_transform(image1) │
│ 116 │ │ text = ' {}'.format(data[self.column_map['text']])

[Claber]

if input just contains 'image' as non-list but text as a list, it runs without errors but the to-be-tokenized text will be '{}'.format(data[self.column_map['text']]). Is it supposed to be so?