Merge branch 'main' of github.com:ander1119/viper into main

cached_code/v+d_df_gpt4_simple/v1.csv

@@ -45,7 +45,7 @@ def execute_command(video, annotation, possible_answers, query):
  big_bad_candidate = character_id
 
  # If a candidate is found and their negative impacts are significant, answer 'yes'
- if big_bad_candidate and max_negative_impacts > len(video_segment.frame_iterator()) * 0.5: # Arbitrary threshold: more than 50% of frames
+ if big_bad_candidate and max_negative_impacts > len(video_segment) * 0.5: # Arbitrary threshold: more than 50% of frames
  answer = ""yes""
  reason = f""The character {big_bad_candidate} caused negative impacts in more than 50% of the frames.""
  else:

vision_models.py

@@ -332,7 +332,7 @@ def forward(self, image, prompt, task='score', return_index=True, negative_categ
 class MaskRCNNModel(BaseModel):
  name = 'maskrcnn'
 
- def __init__(self, gpu_number=1, threshold=config.detect_thresholds.maskrcnn):
+ def __init__(self, gpu_number=0, threshold=config.detect_thresholds.maskrcnn):
  super().__init__(gpu_number)
  # with HiddenPrints('MaskRCNN'):
  obj_detect = torchvision.models.detection.maskrcnn_resnet50_fpn_v2(weights='COCO_V1').to(self.dev)
@@ -429,7 +429,7 @@ def forward(self, image: torch.Tensor, text: List[str], return_labels: bool = Fa
 class GLIPModel(BaseModel):
  name = 'glip'
 
- def __init__(self, model_size='large', gpu_number=1, *args):
+ def __init__(self, model_size='large', gpu_number=0, *args):
  BaseModel.__init__(self, gpu_number)
 
  with contextlib.redirect_stderr(open(os.devnull, "w")): # Do not print nltk_data messages when importing
@@ -910,13 +910,17 @@ def get_summarization(self, prompts) -> list[dict]:
  "content": prompt
  }
  ]
- response = openai.ChatCompletion.create(
- model=self.model,
- messages=message,
- # response_format={"type": "json_object"},
- temperature=self.temperature,
- )
- responses.append(response.choices[0].message.content)
+ try:
+ response = openai.ChatCompletion.create(
+ model=self.model,
+ messages=message,
+ # response_format={"type": "json_object"},
+ temperature=self.temperature,
+ )
+ responses.append(response.choices[0].message.content)
+ except:
+ responses.append("")
+
  return responses
 
  def query_gpt3(self, prompt, model="text-davinci-003", max_tokens=16, logprobs=None, stream=False,
@@ -1394,7 +1398,7 @@ class BLIPModel(BaseModel):
  max_batch_size = 32
  seconds_collect_data = 0.2 # The queue has additionally the time it is executing the previous forward pass
 
- def __init__(self, gpu_number=1, half_precision=config.blip_half_precision,
+ def __init__(self, gpu_number=0, half_precision=config.blip_half_precision,
  blip_v2_model_type=config.blip_v2_model_type):
  super().__init__(gpu_number)