upload refined_40k/test_bench 's json, videos and frames

data/refined_40k/test_bench/data_bad_insf_static.json

@@ -0,0 +1,288 @@
+[
+    {
+        "id": "d700567",
+        "images": [
+            "d700567_00.jpg",
+            "d700567_01.jpg",
+            "d700567_02.jpg",
+            "d700567_03.jpg",
+            "d700567_04.jpg",
+            "d700567_05.jpg",
+            "d700567_06.jpg",
+            "d700567_07.jpg"
+        ],
+        "conversations": [
+            {
+                "from": "human",
+                "value": "\nSuppose you are an expert in judging and evaluating the quality of AI-generated videos, \nplease watch the following frames of a given video and see the text prompt for generating the video, \nthen give scores from 5 different dimensions:\n(1) visual quality: the quality of the video in terms of clearness, resolution, brightness, and color\n(2) temporal consistency, the consistency of objects or humans in video\n(3) dynamic degree, the degree of dynamic changes\n(4) text-to-video alignment, the alignment between the text prompt and the video content\n(5) factual consistency, the consistency of the video content with the common-sense and factual knowledge\n\nFor each dimension, output a number from [1,2,3,4], \nin which '1' means 'Bad', '2' means 'Average', '3' means 'Good', \n'4' means 'Real' or 'Perfect' (the video is like a real video)\nHere is an output example:\nvisual quality: 4\ntemporal consistency: 4\ndynamic degree: 3\ntext-to-video alignment: 1\nfactual consistency: 2\n\nFor this video, the text prompt is \"slow French movie, a person waiting at a cafe at dawn  \",\nall the frames of video are as follows: \n\n<image> <image> <image> <image> <image> <image> <image> <image> \n"
+            },
+            {
+                "from": "gpt",
+                "value": "visual quality: 1\n temporal consistency: 3\n dynamic degree: 1\n text-to-video alignment: 1\n factual consistency: 1\n"
+            }
+        ]
+    },
+    {
+        "id": "d402339",
+        "images": [
+            "d402339_00.jpg",
+            "d402339_01.jpg",
+            "d402339_02.jpg",
+            "d402339_03.jpg",
+            "d402339_04.jpg",
+            "d402339_05.jpg",
+            "d402339_06.jpg",
+            "d402339_07.jpg",
+            "d402339_08.jpg",
+            "d402339_09.jpg",
+            "d402339_10.jpg",
+            "d402339_11.jpg",
+            "d402339_12.jpg",
+            "d402339_13.jpg",
+            "d402339_14.jpg",
+            "d402339_15.jpg"
+        ],
+        "conversations": [
+            {
+                "from": "human",
+                "value": "\nSuppose you are an expert in judging and evaluating the quality of AI-generated videos, \nplease watch the following frames of a given video and see the text prompt for generating the video, \nthen give scores from 5 different dimensions:\n(1) visual quality: the quality of the video in terms of clearness, resolution, brightness, and color\n(2) temporal consistency, the consistency of objects or humans in video\n(3) dynamic degree, the degree of dynamic changes\n(4) text-to-video alignment, the alignment between the text prompt and the video content\n(5) factual consistency, the consistency of the video content with the common-sense and factual knowledge\n\nFor each dimension, output a number from [1,2,3,4], \nin which '1' means 'Bad', '2' means 'Average', '3' means 'Good', \n'4' means 'Real' or 'Perfect' (the video is like a real video)\nHere is an output example:\nvisual quality: 4\ntemporal consistency: 4\ndynamic degree: 3\ntext-to-video alignment: 1\nfactual consistency: 2\n\nFor this video, the text prompt is \"Dynamic introduction of the speakers with a tech background\uff0cExploring the secrets of human marketing, entering a new era of AI live streaming. Film effect.  \",\nall the frames of video are as follows: \n\n<image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> \n"
+            },
+            {
+                "from": "gpt",
+                "value": "visual quality: 1\n temporal consistency: 3\n dynamic degree: 1\n text-to-video alignment: 1\n factual consistency: 1\n"
+            }
+        ]
+    },
+    {
+        "id": "d700935",
+        "images": [
+            "d700935_00.jpg",
+            "d700935_01.jpg",
+            "d700935_02.jpg",
+            "d700935_03.jpg",
+            "d700935_04.jpg",
+            "d700935_05.jpg",
+            "d700935_06.jpg",
+            "d700935_07.jpg"
+        ],
+        "conversations": [
+            {
+                "from": "human",
+                "value": "\nSuppose you are an expert in judging and evaluating the quality of AI-generated videos, \nplease watch the following frames of a given video and see the text prompt for generating the video, \nthen give scores from 5 different dimensions:\n(1) visual quality: the quality of the video in terms of clearness, resolution, brightness, and color\n(2) temporal consistency, the consistency of objects or humans in video\n(3) dynamic degree, the degree of dynamic changes\n(4) text-to-video alignment, the alignment between the text prompt and the video content\n(5) factual consistency, the consistency of the video content with the common-sense and factual knowledge\n\nFor each dimension, output a number from [1,2,3,4], \nin which '1' means 'Bad', '2' means 'Average', '3' means 'Good', \n'4' means 'Real' or 'Perfect' (the video is like a real video)\nHere is an output example:\nvisual quality: 4\ntemporal consistency: 4\ndynamic degree: 3\ntext-to-video alignment: 1\nfactual consistency: 2\n\nFor this video, the text prompt is \"He will encounter various challenges and obstacles on his journey, but he firmly believes that only solving the mystery of the lost city can lead to important discoveries about lost civilizations.  \",\nall the frames of video are as follows: \n\n<image> <image> <image> <image> <image> <image> <image> <image> \n"
+            },
+            {
+                "from": "gpt",
+                "value": "visual quality: 1\n temporal consistency: 3\n dynamic degree: 1\n text-to-video alignment: 1\n factual consistency: 1\n"
+            }
+        ]
+    },
+    {
+        "id": "d401345",
+        "images": [
+            "d401345_00.jpg",
+            "d401345_01.jpg",
+            "d401345_02.jpg",
+            "d401345_03.jpg",
+            "d401345_04.jpg",
+            "d401345_05.jpg",
+            "d401345_06.jpg",
+            "d401345_07.jpg",
+            "d401345_08.jpg",
+            "d401345_09.jpg",
+            "d401345_10.jpg",
+            "d401345_11.jpg",
+            "d401345_12.jpg",
+            "d401345_13.jpg",
+            "d401345_14.jpg",
+            "d401345_15.jpg"
+        ],
+        "conversations": [
+            {
+                "from": "human",
+                "value": "\nSuppose you are an expert in judging and evaluating the quality of AI-generated videos, \nplease watch the following frames of a given video and see the text prompt for generating the video, \nthen give scores from 5 different dimensions:\n(1) visual quality: the quality of the video in terms of clearness, resolution, brightness, and color\n(2) temporal consistency, the consistency of objects or humans in video\n(3) dynamic degree, the degree of dynamic changes\n(4) text-to-video alignment, the alignment between the text prompt and the video content\n(5) factual consistency, the consistency of the video content with the common-sense and factual knowledge\n\nFor each dimension, output a number from [1,2,3,4], \nin which '1' means 'Bad', '2' means 'Average', '3' means 'Good', \n'4' means 'Real' or 'Perfect' (the video is like a real video)\nHere is an output example:\nvisual quality: 4\ntemporal consistency: 4\ndynamic degree: 3\ntext-to-video alignment: 1\nfactual consistency: 2\n\nFor this video, the text prompt is \"tries to catch the colorful ribbons under the Christmas tree.,  \",\nall the frames of video are as follows: \n\n<image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> \n"
+            },
+            {
+                "from": "gpt",
+                "value": "visual quality: 1\n temporal consistency: 3\n dynamic degree: 1\n text-to-video alignment: 1\n factual consistency: 1\n"
+            }
+        ]
+    },
+    {
+        "id": "d701841",
+        "images": [
+            "d701841_00.jpg",
+            "d701841_01.jpg",
+            "d701841_02.jpg",
+            "d701841_03.jpg",
+            "d701841_04.jpg",
+            "d701841_05.jpg",
+            "d701841_06.jpg",
+            "d701841_07.jpg"
+        ],
+        "conversations": [
+            {
+                "from": "human",
+                "value": "\nSuppose you are an expert in judging and evaluating the quality of AI-generated videos, \nplease watch the following frames of a given video and see the text prompt for generating the video, \nthen give scores from 5 different dimensions:\n(1) visual quality: the quality of the video in terms of clearness, resolution, brightness, and color\n(2) temporal consistency, the consistency of objects or humans in video\n(3) dynamic degree, the degree of dynamic changes\n(4) text-to-video alignment, the alignment between the text prompt and the video content\n(5) factual consistency, the consistency of the video content with the common-sense and factual knowledge\n\nFor each dimension, output a number from [1,2,3,4], \nin which '1' means 'Bad', '2' means 'Average', '3' means 'Good', \n'4' means 'Real' or 'Perfect' (the video is like a real video)\nHere is an output example:\nvisual quality: 4\ntemporal consistency: 4\ndynamic degree: 3\ntext-to-video alignment: 1\nfactual consistency: 2\n\nFor this video, the text prompt is \"animation style, miyazaki, a big window, a rain throw a window, by rainy day, with view of sea,  \",\nall the frames of video are as follows: \n\n<image> <image> <image> <image> <image> <image> <image> <image> \n"
+            },
+            {
+                "from": "gpt",
+                "value": "visual quality: 1\n temporal consistency: 3\n dynamic degree: 1\n text-to-video alignment: 1\n factual consistency: 1\n"
+            }
+        ]
+    },
+    {
+        "id": "d500552",
+        "images": [
+            "d500552_00.jpg",
+            "d500552_01.jpg",
+            "d500552_02.jpg",
+            "d500552_03.jpg",
+            "d500552_04.jpg",
+            "d500552_05.jpg",
+            "d500552_06.jpg",
+            "d500552_07.jpg",
+            "d500552_08.jpg",
+            "d500552_09.jpg",
+            "d500552_10.jpg",
+            "d500552_11.jpg",
+            "d500552_12.jpg",
+            "d500552_13.jpg",
+            "d500552_14.jpg",
+            "d500552_15.jpg"
+        ],
+        "conversations": [
+            {
+                "from": "human",
+                "value": "\nSuppose you are an expert in judging and evaluating the quality of AI-generated videos, \nplease watch the following frames of a given video and see the text prompt for generating the video, \nthen give scores from 5 different dimensions:\n(1) visual quality: the quality of the video in terms of clearness, resolution, brightness, and color\n(2) temporal consistency, the consistency of objects or humans in video\n(3) dynamic degree, the degree of dynamic changes\n(4) text-to-video alignment, the alignment between the text prompt and the video content\n(5) factual consistency, the consistency of the video content with the common-sense and factual knowledge\n\nFor each dimension, output a number from [1,2,3,4], \nin which '1' means 'Bad', '2' means 'Average', '3' means 'Good', \n'4' means 'Real' or 'Perfect' (the video is like a real video)\nHere is an output example:\nvisual quality: 4\ntemporal consistency: 4\ndynamic degree: 3\ntext-to-video alignment: 1\nfactual consistency: 2\n\nFor this video, the text prompt is \"man on a blue and pink planet  \",\nall the frames of video are as follows: \n\n<image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> \n"
+            },
+            {
+                "from": "gpt",
+                "value": "visual quality: 1\n temporal consistency: 3\n dynamic degree: 1\n text-to-video alignment: 1\n factual consistency: 1\n"
+            }
+        ]
+    },
+    {
+        "id": "d500525",
+        "images": [
+            "d500525_00.jpg",
+            "d500525_01.jpg",
+            "d500525_02.jpg",
+            "d500525_03.jpg",
+            "d500525_04.jpg",
+            "d500525_05.jpg",
+            "d500525_06.jpg",
+            "d500525_07.jpg",
+            "d500525_08.jpg",
+            "d500525_09.jpg",
+            "d500525_10.jpg",
+            "d500525_11.jpg",
+            "d500525_12.jpg",
+            "d500525_13.jpg",
+            "d500525_14.jpg",
+            "d500525_15.jpg"
+        ],
+        "conversations": [
+            {
+                "from": "human",
+                "value": "\nSuppose you are an expert in judging and evaluating the quality of AI-generated videos, \nplease watch the following frames of a given video and see the text prompt for generating the video, \nthen give scores from 5 different dimensions:\n(1) visual quality: the quality of the video in terms of clearness, resolution, brightness, and color\n(2) temporal consistency, the consistency of objects or humans in video\n(3) dynamic degree, the degree of dynamic changes\n(4) text-to-video alignment, the alignment between the text prompt and the video content\n(5) factual consistency, the consistency of the video content with the common-sense and factual knowledge\n\nFor each dimension, output a number from [1,2,3,4], \nin which '1' means 'Bad', '2' means 'Average', '3' means 'Good', \n'4' means 'Real' or 'Perfect' (the video is like a real video)\nHere is an output example:\nvisual quality: 4\ntemporal consistency: 4\ndynamic degree: 3\ntext-to-video alignment: 1\nfactual consistency: 2\n\nFor this video, the text prompt is \"5 seconds video,  a bottle falling from a table, while crushing on the ground pieces of glass become colorfully candies, hyper realistic video, dark mood  \",\nall the frames of video are as follows: \n\n<image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> \n"
+            },
+            {
+                "from": "gpt",
+                "value": "visual quality: 1\n temporal consistency: 3\n dynamic degree: 1\n text-to-video alignment: 1\n factual consistency: 1\n"
+            }
+        ]
+    },
+    {
+        "id": "d401859",
+        "images": [
+            "d401859_00.jpg",
+            "d401859_01.jpg",
+            "d401859_02.jpg",
+            "d401859_03.jpg",
+            "d401859_04.jpg",
+            "d401859_05.jpg",
+            "d401859_06.jpg",
+            "d401859_07.jpg",
+            "d401859_08.jpg",
+            "d401859_09.jpg",
+            "d401859_10.jpg",
+            "d401859_11.jpg",
+            "d401859_12.jpg",
+            "d401859_13.jpg",
+            "d401859_14.jpg",
+            "d401859_15.jpg"
+        ],
+        "conversations": [
+            {
+                "from": "human",
+                "value": "\nSuppose you are an expert in judging and evaluating the quality of AI-generated videos, \nplease watch the following frames of a given video and see the text prompt for generating the video, \nthen give scores from 5 different dimensions:\n(1) visual quality: the quality of the video in terms of clearness, resolution, brightness, and color\n(2) temporal consistency, the consistency of objects or humans in video\n(3) dynamic degree, the degree of dynamic changes\n(4) text-to-video alignment, the alignment between the text prompt and the video content\n(5) factual consistency, the consistency of the video content with the common-sense and factual knowledge\n\nFor each dimension, output a number from [1,2,3,4], \nin which '1' means 'Bad', '2' means 'Average', '3' means 'Good', \n'4' means 'Real' or 'Perfect' (the video is like a real video)\nHere is an output example:\nvisual quality: 4\ntemporal consistency: 4\ndynamic degree: 3\ntext-to-video alignment: 1\nfactual consistency: 2\n\nFor this video, the text prompt is \"A kitten and a hummingbird play in a garden  \",\nall the frames of video are as follows: \n\n<image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> \n"
+            },
+            {
+                "from": "gpt",
+                "value": "visual quality: 1\n temporal consistency: 3\n dynamic degree: 1\n text-to-video alignment: 1\n factual consistency: 1\n"
+            }
+        ]
+    },
+    {
+        "id": "d500773",
+        "images": [
+            "d500773_00.jpg",
+            "d500773_01.jpg",
+            "d500773_02.jpg",
+            "d500773_03.jpg",
+            "d500773_04.jpg",
+            "d500773_05.jpg",
+            "d500773_06.jpg",
+            "d500773_07.jpg",
+            "d500773_08.jpg",
+            "d500773_09.jpg",
+            "d500773_10.jpg",
+            "d500773_11.jpg",
+            "d500773_12.jpg",
+            "d500773_13.jpg",
+            "d500773_14.jpg",
+            "d500773_15.jpg"
+        ],
+        "conversations": [
+            {
+                "from": "human",
+                "value": "\nSuppose you are an expert in judging and evaluating the quality of AI-generated videos, \nplease watch the following frames of a given video and see the text prompt for generating the video, \nthen give scores from 5 different dimensions:\n(1) visual quality: the quality of the video in terms of clearness, resolution, brightness, and color\n(2) temporal consistency, the consistency of objects or humans in video\n(3) dynamic degree, the degree of dynamic changes\n(4) text-to-video alignment, the alignment between the text prompt and the video content\n(5) factual consistency, the consistency of the video content with the common-sense and factual knowledge\n\nFor each dimension, output a number from [1,2,3,4], \nin which '1' means 'Bad', '2' means 'Average', '3' means 'Good', \n'4' means 'Real' or 'Perfect' (the video is like a real video)\nHere is an output example:\nvisual quality: 4\ntemporal consistency: 4\ndynamic degree: 3\ntext-to-video alignment: 1\nfactual consistency: 2\n\nFor this video, the text prompt is \"kung fu cats vs dog in a football match video duration one minute  \",\nall the frames of video are as follows: \n\n<image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> \n"
+            },
+            {
+                "from": "gpt",
+                "value": "visual quality: 1\n temporal consistency: 3\n dynamic degree: 1\n text-to-video alignment: 1\n factual consistency: 1\n"
+            }
+        ]
+    },
+    {
+        "id": "d400053",
+        "images": [
+            "d400053_00.jpg",
+            "d400053_01.jpg",
+            "d400053_02.jpg",
+            "d400053_03.jpg",
+            "d400053_04.jpg",
+            "d400053_05.jpg",
+            "d400053_06.jpg",
+            "d400053_07.jpg",
+            "d400053_08.jpg",
+            "d400053_09.jpg",
+            "d400053_10.jpg",
+            "d400053_11.jpg",
+            "d400053_12.jpg",
+            "d400053_13.jpg",
+            "d400053_14.jpg",
+            "d400053_15.jpg"
+        ],
+        "conversations": [
+            {
+                "from": "human",
+                "value": "\nSuppose you are an expert in judging and evaluating the quality of AI-generated videos, \nplease watch the following frames of a given video and see the text prompt for generating the video, \nthen give scores from 5 different dimensions:\n(1) visual quality: the quality of the video in terms of clearness, resolution, brightness, and color\n(2) temporal consistency, the consistency of objects or humans in video\n(3) dynamic degree, the degree of dynamic changes\n(4) text-to-video alignment, the alignment between the text prompt and the video content\n(5) factual consistency, the consistency of the video content with the common-sense and factual knowledge\n\nFor each dimension, output a number from [1,2,3,4], \nin which '1' means 'Bad', '2' means 'Average', '3' means 'Good', \n'4' means 'Real' or 'Perfect' (the video is like a real video)\nHere is an output example:\nvisual quality: 4\ntemporal consistency: 4\ndynamic degree: 3\ntext-to-video alignment: 1\nfactual consistency: 2\n\nFor this video, the text prompt is \"man sitting on bedin bedroom,  hands in the air. hundreds of various printed pillows falling down from above  \",\nall the frames of video are as follows: \n\n<image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> <image> \n"
+            },
+            {
+                "from": "gpt",
+                "value": "visual quality: 1\n temporal consistency: 3\n dynamic degree: 1\n text-to-video alignment: 1\n factual consistency: 1\n"
+            }
+        ]
+    }
+]