import os import json import collections from pathlib import Path import numpy as np import torch from data.data_utils import SQA3DAnswer from evaluator.build import EVALUATOR_REGISTRY import re @EVALUATOR_REGISTRY.register() class SQA3DEval(): # 0: what, 1: is, 2: how, 3: can, 4: which, 5: others def __init__(self, cfg, task_name): self.eval_dict = { 'target_metric': [], 'ans1_acc': [], 'ans10_acc': [], 'non_color_ans1_acc': [], 'non_color_ans10_acc': [], 'type0_acc': [], 'type1_acc': [], 'type2_acc': [], 'type0_acc': [], 'type1_acc': [], 'type2_acc': [], 'type3_acc': [], 'type4_acc': [], 'type5_acc': [] } # run self.total_count = 0 self.non_color_total_count = 0 self.type_count = { 'type0_count': 1e-10, 'type1_count': 1e-10, 'type2_count': 1e-10, 'type3_count': 1e-10, 'type4_count': 1e-10, 'type5_count': 1e-10 } self.best_result = -np.inf self.base_dir = cfg.data.scan_family_base answer_data = json.load( open(os.path.join(self.base_dir, 'annotations/sqa_task/answer_dict.json'), encoding='utf-8') )[0] color_terms = [ "orange", "pink", "maroon", "grey", "gray", "purple", "red", "yellow", "brown", "blue", "green", "silver", "gold", "tan", "turquoise", "beige", "white", "black", "chocolate", "multicolored", "black and red", "yellow and orange", "black white", "light brown", "dark brown", "balck", "whitec" ] # Collect all color-related answers with their ids self.color_ids = [answer_data[name] for name in color_terms] answer_counter = [] for data in answer_data.keys(): answer_counter.append(data) answer_counter = collections.Counter(sorted(answer_counter)) answer_cands = answer_counter.keys() self.answer_vocab = SQA3DAnswer(answer_cands) self.save = cfg.eval.save if self.save: self.eval_results = [] self.save_dir = Path(cfg.exp_dir) / "eval_results" / task_name self.save_dir.mkdir(parents=True, exist_ok=True) def update(self, data_dict): metrics = self.batch_metrics(data_dict) batch_count = metrics['total_count'] self.total_count += batch_count non_color_batch_count = metrics['non_color_total_count'] self.non_color_total_count += non_color_batch_count for key in metrics: if 'type' in key and 'count' in key: self.type_count[key] += metrics[key] if self.save: for i in range(metrics["total_count"]): self.eval_results.append({ # vision "source": data_dict['source'][i], "scan_id": data_dict['scan_id'][i], "anchor": data_dict['anchor_locs'][i], 'anchor_ort': data_dict['anchor_orientation'][i], # language "instruction": data_dict['prompt_after_obj'][i], "response_gt": data_dict['answer_list'][i].split('[answer_seq]'), "response_pred": data_dict['output_text'][i] }) # save eval dict for key in self.eval_dict.keys(): if 'type' in key: self.eval_dict[key].append(float(metrics[key]) * metrics['type' + key[4] + '_count']) elif 'non_color' not in key: self.eval_dict[key].append(float(metrics[key]) * batch_count) else: self.eval_dict[key].append(float(metrics[key]) * non_color_batch_count) def batch_metrics(self, data_dict): metrics = {} # ans choice_1 = data_dict['answer_scores'].argmax(dim=-1) choice_10 = torch.topk(data_dict['answer_scores'].detach(), 10, -1)[1] correct1 = 0 correct10 = 0 non_color_correct1 = 0 non_color_correct10 = 0 total_non_color = 0 correct_type = {0: 0, 1: 0, 2: 0, 3: 0, 4: 0, 5: 0} count_type = {0: 1e-10, 1: 1e-10, 2: 1e-10, 3: 1e-10, 4: 1e-10, 5: 1e-10} for i in range(data_dict['answer_label'].shape[0]): gt_id = data_dict['answer_label'][i].argmax().item() if gt_id not in self.color_ids: total_non_color += 1 count_type[data_dict['sqa_type'][i].item()] += 1 if data_dict['answer_label'][i, choice_1[i]] == 1: if choice_1[i] not in self.color_ids: non_color_correct1 += 1 correct1 += 1 correct_type[data_dict['sqa_type'][i].item()] += 1 for j in range(10): if data_dict['answer_label'][i, choice_10[i, j]] == 1: if choice_10[i,j] not in self.color_ids: non_color_correct10 += 1 correct10 += 1 break metrics['ans1_acc'] = correct1 / float(len(choice_1)) metrics['ans10_acc'] = correct10 / float(len(choice_1)) metrics['non_color_ans1_acc'] = non_color_correct1 / float(total_non_color) metrics['non_color_ans10_acc'] = non_color_correct10 / float(total_non_color) # question type acc for key in count_type.keys(): metrics['type' + str(key) + '_acc'] = correct_type[key] / count_type[key] metrics['type' + str(key) + '_count'] = count_type[key] metrics['target_metric'] = metrics['ans1_acc'] metrics["total_count"] = data_dict["answer_scores"].shape[0] metrics["non_color_total_count"] = total_non_color return metrics def reset(self): for key in self.eval_dict.keys(): self.eval_dict[key] = [] self.total_count = 0 self.non_color_total_count = 0 self.type_count = { 'type0_count': 1e-10, 'type1_count': 1e-10, 'type2_count': 1e-10, 'type3_count': 1e-10, 'type4_count': 1e-10, 'type5_count': 1e-10 } if self.save: self.eval_results = [] def record(self, split='val'): # record for k, v in self.eval_dict.items(): if k == "answer_top10": continue if 'type' in k: self.eval_dict[k] = sum(v) / self.type_count['type' + k[4] + '_count'] elif 'non_color' not in k: self.eval_dict[k] = sum(v) / self.total_count print(k, 'overall', sum(v), self.total_count) elif 'non_color' in k: self.eval_dict[k] = sum(v) / self.non_color_total_count print(k, 'non_color', sum(v), self.non_color_total_count) if self.eval_dict["target_metric"] > self.best_result: is_best = True self.best_result = self.eval_dict["target_metric"] else: is_best = False if self.save and (is_best or split == 'test'): torch.save(self.eval_results, str(self.save_dir / 'results.pt')) return is_best, self.eval_dict