fixed imports (#19)

Author: yeshaokai

llavaction/action/benchmark.py

@@ -1,7 +1,7 @@
 # benchmark gpt-4o on avion_mcq_top5_500
 # benchmark gpt-4o on tim_mcq_top5_500
 # benchmark gpt-4o on random_mcq_top5_500
-from llava.action.chatgpt_utils import GPTInferenceAnnotator
+from llavaction.action.chatgpt_utils import GPTInferenceAnnotator
 import glob
 import json
 import os

llavaction/action/chatgpt_utils.py

@@ -6,11 +6,11 @@
 import json
 from concurrent.futures import ProcessPoolExecutor
 from tqdm import tqdm
-from llava.action.utils import AvionMultiChoiceGenerator as ActionMultiChoiceGenerator
-from llava.action.utils import RandomMultiChoiceGenerator
-from llava.action.utils import avion_video_loader, avion_video_render_loader, generate_label_map
-from llava.action.dataset import datetime2sec
-from llava.action.ek_eval import process_raw_pred
+from llavaction.action.utils import AvionMultiChoiceGenerator as ActionMultiChoiceGenerator
+from llavaction.action.utils import RandomMultiChoiceGenerator
+from llavaction.action.utils import avion_video_loader, avion_video_render_loader, generate_label_map
+from llavaction.action.dataset import datetime2sec
+from llavaction.action.ek_eval import process_raw_pred
 import csv
 import copy 
 import torch
@@ -543,7 +543,7 @@ def predict_images(self, images, parsed_item):
         """
         Predict the action from the images
         """
-        from llava.action.utils import format_task_related_prompt
+        from llavaction.action.utils import format_task_related_prompt
         options = parsed_item['options']
         start_second = 0
         end_second = parsed_item['end_second'] - parsed_item['start_second']

llavaction/action/dataset.py

@@ -6,8 +6,8 @@
 import torch
 import decord
 from pathlib import Path
-from llava.action.utils import  AvionMultiChoiceGenerator,  RandomMultiChoiceGenerator, avion_video_loader, EK100_frame_loader
-from llava.action.prediction_analysis import PredictionAnalysis
+from llavaction.action.utils import  AvionMultiChoiceGenerator,  RandomMultiChoiceGenerator, avion_video_loader, EK100_frame_loader
+from llavaction.action.prediction_analysis import PredictionAnalysis
 import torch.distributed as dist
 
 def datetime2sec(str):

llavaction/action/ek_eval.py

@@ -7,14 +7,14 @@
 import sys
 import os
 import numpy as np
-from llava.action.llava_inference import llava_inference
+from llavaction.action.llava_inference import llava_inference
 import json
 import logging
-from llava.utils import rank0_print
-from llava.action.utils import generate_label_map
+from llavaction.utils import rank0_print
+from llavaction.action.utils import generate_label_map
 from collections import Counter 
 import torch.distributed as dist
-from llava.action.dataset import VideoMultiChoiceDataset
+from llavaction.action.dataset import VideoMultiChoiceDataset
 import torchvision.io as io
 import re
 
@@ -138,7 +138,7 @@ def prepare_llava(pretrained, use_flash_attention = True):
 
     import warnings
     warnings.filterwarnings("ignore")
-    from llava.model.builder import load_pretrained_model    
+    from llavaction.model.builder import load_pretrained_model    
     model_name = "llava_qwen"
 
     device_map = "auto"
@@ -359,7 +359,7 @@ def collate_fn(batch):
     lookup_table = None
     meta_data = None
     if eval_args.learn_neighbor_actions:
-        from llava.action.generate_interval_pred import  get_lookup_dict
+        from llavaction.action.generate_interval_pred import  get_lookup_dict
         if eval_args.test_type.startswith('temporal_cot'):
             lookup_table = get_lookup_dict(eval_args.val_metadata, 
                                            eval_args.action_representation,

llavaction/action/generate_description.py

@@ -6,11 +6,11 @@
 import sys
 import numpy as np
 sys.path[0] = os.path.dirname(os.path.dirname(sys.path[0]))
-from llava.action.utils import generate_label_map, RandomMultiChoiceGenerator, AvionMultiChoiceGenerator, format_llava_prompt, remove_sub_nouns
-from llava.action.dataset import datetime2sec
+from llavaction.action.utils import generate_label_map, RandomMultiChoiceGenerator, AvionMultiChoiceGenerator, format_llava_prompt, remove_sub_nouns
+from llavaction.action.dataset import datetime2sec
 from pathlib import Path
-from llava.action.utils import hand_obj_ann_loader
-from llava.action.generate_interval_pred import build_uid_pad_dict
+from llavaction.action.utils import hand_obj_ann_loader
+from llavaction.action.generate_interval_pred import build_uid_pad_dict
 import ast
 
 def generate_train_ann(ann_file, labels, mapping_vn2narration, mapping_vn2act, verb_maps, noun_maps, gen_type = 'naive', prediction_path = '', n_options = 5,

llavaction/action/generate_interval_pred.py

@@ -10,7 +10,7 @@
 from pathlib import Path
 from tqdm import tqdm
 from collections import Counter
-from llava.action.utils import generate_label_map
+from llavaction.action.utils import generate_label_map
 
 def datetime2sec(str):
     hh, mm, ss = str.split(':')

llavaction/action/generate_temporal_detection_data.py

@@ -1,8 +1,8 @@
 
 import csv
-from llava.action.dataset import datetime2sec
+from llavaction.action.dataset import datetime2sec
 import random
-from llava.action.utils import generate_label_map
+from llavaction.action.utils import generate_label_map
 from pathlib import Path
 import json
 

llavaction/action/llava_inference.py

@@ -1,12 +1,12 @@
-from llava.mm_utils import get_model_name_from_path, process_images, tokenizer_image_token
-from llava.constants import IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN, DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN, IGNORE_INDEX
-from llava.conversation import conv_templates, SeparatorStyle
+from llavaction.mm_utils import get_model_name_from_path, process_images, tokenizer_image_token
+from llavaction.constants import IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN, DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN, IGNORE_INDEX
+from llavaction.conversation import conv_templates, SeparatorStyle
 
 import torch
 import numpy as np
 import copy
-from llava.action.utils import format_llava_prompt
-from llava.utils import rank0_print
+from llavaction.action.utils import format_llava_prompt
+from llavaction.utils import rank0_print
 
 
 def llava_inference(

llavaction/action/make_visualizations.py

@@ -9,11 +9,11 @@
 
 Note that in each inference, we should be able to pick the corresponding prompt and checkpoint folder
 """
-from llava.action.utils import generate_label_map  
+from llavaction.action.utils import generate_label_map  
 
 from pathlib import Path
-from llava.action.utils import AvionMultiChoiceGenerator as ActionMultiChoiceGenerator
-from llava.action.llava_inference import llava_inference
+from llavaction.action.utils import AvionMultiChoiceGenerator as ActionMultiChoiceGenerator
+from llavaction.action.llava_inference import llava_inference
 import json
 import cv2
 # root = '/data/EK100/EK100_320p_15sec_30fps_libx264'
@@ -40,7 +40,7 @@ def visualize_with_random(n_samples, offset = 0, question_type = 'mc_'):
     """
     Here we should test gpt-4o, gpt-4o-mini with different prompts
     """
-    from llava.action.chatgpt_utils import GPTInferenceAnnotator
+    from llavaction.action.chatgpt_utils import GPTInferenceAnnotator
     inferencer = GPTInferenceAnnotator(gpt_model,
                                        root,
                                        annotation_file,
@@ -60,7 +60,7 @@ def visualize_with_gpt_with_tim(n_samples, offset = 0, question_type = 'mc_'):
     """
     Here we should test gpt-4o, gpt-4o-mini with different prompts
     """
-    from llava.action.chatgpt_utils import GPTInferenceAnnotator
+    from llavaction.action.chatgpt_utils import GPTInferenceAnnotator
     inferencer = GPTInferenceAnnotator(gpt_model,
                                        root,
                                        annotation_file,
@@ -81,7 +81,7 @@ def visualize_with_gpt_with_avion(n_samples, offset = 0, question_type = 'mc_'):
     """
     Here we should test gpt-4o, gpt-4o-mini with different prompts
     """
-    from llava.action.chatgpt_utils import GPTInferenceAnnotator
+    from llavaction.action.chatgpt_utils import GPTInferenceAnnotator
     inferencer = GPTInferenceAnnotator(gpt_model,
                                        root,
                                        annotation_file,
@@ -100,7 +100,7 @@ def visualize_with_gpt_with_avion(n_samples, offset = 0, question_type = 'mc_'):
 
 def search_option_data_by_uid(uid, anno_file, gen_type = 'tim'):
     import csv
-    from llava.action.dataset import datetime2sec
+    from llavaction.action.dataset import datetime2sec
     csv_reader = csv.reader(open(anno_file, 'r'))
     _ = next(csv_reader) # skip the header
     query_vid_path = '_'.join(uid.split('_')[:2]).replace('-', '/')
@@ -166,7 +166,7 @@ def save_visualization(vis_folder, frames, uid):
     video_out.release()
 
 def visualize_with_uid(data_root, uid, out_folder):
-    from llava.action.utils import avion_video_loader
+    from llavaction.action.utils import avion_video_loader
 
     vid_path = '_'.join(uid.split('_')[:2]).replace('-', '/')
     start_timestamp, end_timestamp = uid.split('_')[2:]
@@ -193,12 +193,12 @@ def visualize_with_uid(data_root, uid, out_folder):
 def visualize_with_llava(pretrained_path, uid, question_type, gen_type):
     """    
     """
-    from llava.action.ek_eval import prepare_llava
-    from llava.action.dataset import VideoMultiChoiceDataset
+    from llavaction.action.ek_eval import prepare_llava
+    from llavaction.action.dataset import VideoMultiChoiceDataset
 
     import torch
 
-    from llava.action.utils import avion_video_loader
+    from llavaction.action.utils import avion_video_loader
     val_metadata = '/data/shaokai/epic-kitchens-100-annotations/EPIC_100_validation.csv'
 
     gpu_val_transform_ls = []

llavaction/action/prediction_analysis.py

@@ -2,7 +2,7 @@
 import glob
 import os
 import numpy as np
-from llava.action.utils import generate_label_map
+from llavaction.action.utils import generate_label_map
 from tqdm import tqdm
 class PredictionAnalysis:
     """
@@ -94,7 +94,7 @@ def retrieval_accuracy(self, search_type = 'official'):
         self.load()
         llava_preds = [v['llava_pred'] for k, v in self.data.items()]
         gts = [v['gt_name'] for k, v in self.data.items()]
-        from llava.action.retrieval import SemanticRetriever
+        from llavaction.action.retrieval import SemanticRetriever
         retrieval = SemanticRetriever()
         retrieval.load_from_file('embeddings')