To know dataset format

README.md

@@ -182,3 +182,160 @@ If you find **LIBERO** to be useful in your own research, please consider citing
 |------------------|-------------------------------------------------------------------------------------------------------------------------------------|
 | Codebase         | [MIT License](LICENSE)                                                                                                                      |
 | Datasets         | [Creative Commons Attribution 4.0 International (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/legalcode)                 |
+
+# Dataset
+key list:
+📊 data/demo_0/actions
+📊 data/demo_0/dones
+📊 data/demo_0/obs/agentview_rgb
+📊 data/demo_0/obs/ee_ori
+📊 data/demo_0/obs/ee_pos
+📊 data/demo_0/obs/ee_states
+📊 data/demo_0/obs/eye_in_hand_rgb
+📊 data/demo_0/obs/gripper_states
+📊 data/demo_0/obs/joint_states
+📊 data/demo_0/rewards
+📊 data/demo_0/robot_states
+📊 data/demo_0/states
+📊 data/demo_1/actions
+📊 data/demo_1/dones
+📊 data/demo_1/obs/agentview_rgb
+📊 data/demo_1/obs/ee_ori
+📊 data/demo_1/obs/ee_pos
+📊 data/demo_1/obs/ee_states
+📊 data/demo_1/obs/eye_in_hand_rgb
+📊 data/demo_1/obs/gripper_states
+📊 data/demo_1/obs/joint_states
+📊 data/demo_1/rewards
+📊 data/demo_1/robot_states
+📊 data/demo_1/states
+
+detail:
+📊 数据集: data/demo_0/actions
+  Shape: (148, 7) ✅
+  Dtype: float64 ✅
+  Data (first 1 elements): [[ 0.24642857 -0.51696429  0.01607143  0.          0.01821429 -0.
+  -1.        ]] 📈
+
+📊 指定key对应的数据集: data/demo_0/dones
+  Shape: (148,) ✅
+  Dtype: uint8 ✅
+  Data (first 1 elements): [0] 📈
+
+📊 指定key对应的数据集: data/demo_0/obs/agentview_rgb
+  Shape: (148, 128, 128, 3) ✅
+  Dtype: uint8 ✅
+  Data (first 1 elements): [[[[172 164 159]
+   [176 167 163]
+   [175 167 160]
+   ...
+   [199 191 182]
+   [198 189 180]
+   [199 191 182]]
+
+  [[135 132 128]
+   [134 132 128]
+   [134 132 128]
+   ...
+   [138 136 131]
+   [138 136 131]
+   [138 136 132]]]] 📈
+
+📊 指定key对应的数据集: data/demo_0/obs/ee_ori
+  Shape: (148, 3) ✅
+  Dtype: float64 ✅
+  Data (first 1 elements): [[ 3.13415876 -0.04464834 -0.06500138]]
+
+📊 指定key对应的数据集: data/demo_0/obs/ee_pos
+  Shape: (148, 3) ✅
+  Dtype: float64 ✅
+  Data (first 1 elements): [[-0.13461222 -0.03159367  0.26087246]]
+
+📊 指定key对应的数据集: data/demo_0/obs/ee_states
+  Shape: (148, 6) ✅
+  Dtype: float64 ✅
+  Data (first 1 elements): [[-0.13461222 -0.03159367  0.26087246  3.13415876 -0.04464834 -0.06500138]] 📈
+
+📊 指定key对应的数据集: data/demo_0/obs/eye_in_hand_rgb
+  Shape: (148, 128, 128, 3) ✅
+  Dtype: uint8 ✅
+  Data (first 1 elements): [[[[ 57  57  57]
+   [ 57  57  57]
+   [ 57  57  57]
+   ...
+   [ 56  56  56]
+   [ 56  56  56]
+   [ 56  56  56]]
+
+  [[192 182 172]
+   [195 185 175]
+   [203 193 183]
+   ...
+   [102  63  55]
+   [103  85  82]
+   [103  86  84]]]] 📈
+
+📊 指定key对应的数据集: data/demo_0/obs/gripper_states
+  Shape: (148, 2) ✅
+  Dtype: float64 ✅
+  Data (first 1 elements): [[ 0.03610362 -0.03629859]] 📈
+
+📊 指定key对应的数据集: data/demo_0/obs/joint_states
+  Shape: (148, 7) ✅
+  Dtype: float64 ✅
+  Data (first 1 elements): [[-0.01647415 -0.12439009 -0.04895048 -2.40863181 -0.01870208  2.24320355
+   0.7614374 ]] 📈
+
+📊 指定key对应的数据集: data/demo_0/rewards
+  Shape: (148,) ✅
+  Dtype: uint8 ✅
+  Data (first 1 elements): [0] 📈
+
+只有抓到时会奖励：
+📊 指定key对应的数据集: data/demo_0/rewards
+  Shape: (148,) ✅
+  Dtype: uint8 ✅
+  Data (first 1 elements): [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
+ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
+ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
+ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1] 📈
+
+📊 指定key对应的数据集: data/demo_0/robot_states
+  Shape: (148, 9) ✅
+  Dtype: float64 ✅
+  Data (first 1 elements): [[ 0.03610362 -0.03629859 -0.13461222 -0.03159367  0.26087246  0.99967843
+  -0.01424114 -0.02073299  0.00322098]] 📈
+
+0.99967843 -0.01424114 -0.02073299  0.00322098 ❗不清楚对应的是哪里
+
+📊 指定key对应的数据集: data/demo_0/states
+  Shape: (148, 110) ✅
+  Dtype: float64 ✅
+  Data (first 1 elements): [[ 2.50000000e-01 -1.39058727e-02 -1.32895206e-01 -3.95190726e-02
+  -2.41748166e+00 -1.56499662e-02  2.24342081e+00  7.70441485e-01
+   3.39412254e-02 -3.41246158e-02 -1.19341153e-01 -2.39844686e-01
+   3.83793457e-02 -2.26022130e-03  2.27541573e-03  7.07695320e-01
+   7.06510473e-01 -1.43712746e-02  2.45927217e-01 -4.58262675e-03
+   7.07104752e-01 -1.69384343e-03  1.69371900e-03  7.07104753e-01
+   4.99867361e-02 -9.99780001e-02  9.14421729e-02  5.00114670e-01
+   4.99884733e-01  5.00043808e-01  4.99956758e-01 -1.51981690e-01
+   5.50623211e-02  8.92147605e-03  1.60411515e-10 -3.16300943e-07
+  -1.06977855e-06  1.00000000e+00  9.55577769e-02 -1.98123762e-01
+   6.69983663e-02  4.99972463e-01  5.00027595e-01  4.99983556e-01
+   5.00016384e-01  1.50242278e-01  3.00555488e-02  3.83998685e-02
+  -2.29096800e-03  2.29130000e-03  7.07626522e-01  7.06579229e-01
+  -1.99379192e-01 -7.87157716e-02  8.69543094e-03  8.87495516e-10
+  -4.17292295e-07 -2.42161980e-06  1.00000000e+00 -4.50482326e-02
+   1.55016014e-01 -1.65641087e-01  1.61861455e-01 -5.95531864e-02
+  -1.20304390e-02 -1.56137187e-01  5.76449863e-02 -5.71816273e-02
+  -3.19849468e-05  3.20957564e-06  9.09489769e-04  8.99028765e-05
+  -2.15092962e-06 -8.43101306e-04  8.81460714e-11 -7.43320061e-08
+   1.86903244e-05 -1.93300596e-08  1.62516811e-05 -8.34625155e-11
+  -4.66691074e-05  7.71475538e-05 -9.80411604e-01 -1.91370102e-04
+   2.94757962e-06 -3.58290715e-04 -3.41173844e-12 -3.28112589e-12
+   9.13729709e-08 -4.48594712e-10  4.05903512e-10  1.04577420e-13
+  -8.44980612e-06  5.70231207e-05 -3.45229732e-01 -1.41749996e-04
+   1.01964102e-08 -9.34021809e-04  1.53355879e-09  3.00852655e-09
+   4.82651896e-08  7.90797893e-08 -4.52425251e-10  4.04379947e-08
+  -9.63592933e-11 -1.90983833e-11  3.10686229e-06 -1.10637040e-08
+   1.20000589e-08  8.78974798e-12]] 📈

custom/check_dataset_format_test.py

@@ -0,0 +1,117 @@
+import h5py
+import numpy as np
+
+def explore_hdf5(file_path):
+    """
+    查看HDF5文件的结构和数据集内容
+    """
+    try:
+        # ✅ 打开HDF5文件，'r'表示只读模式
+        with h5py.File(file_path, 'r') as f:
+            print("📂 HDF5文件结构:")
+
+            # ✅ 递归函数，用于打印文件中的组和数据集
+            def print_structure(name, obj):
+                indent = '  ' * name.count('/')
+                if isinstance(obj, h5py.Group):
+                    print(f"{indent}📁 Group: {name}")
+                elif isinstance(obj, h5py.Dataset):
+                    print(f"{indent}📊 Dataset: {name}")
+                    print(f"{indent}  Shape: {obj.shape}")
+                    print(f"{indent}  Dtype: {obj.dtype}")
+                    # 如果数据集较小，打印部分数据
+                    if obj.size < 10:
+                        print(f"{indent}  Data: {obj[()]}")
+                    else:
+                        print(f"{indent}  Data (first 1 elements): {obj[:1]}")
+
+            # ✅ 遍历文件结构
+            f.visititems(print_structure)
+
+    except FileNotFoundError:
+        print("❌ 文件不存在，请检查文件路径！")
+    except Exception as e:
+        print(f"❌ 发生错误: {str(e)}")
+
+
+def print_dataset_keys(file_path):
+    """
+    打印HDF5文件中所有数据集的key（名称路径）
+    """
+    try:
+        # ✅ 打开HDF5文件，'r'表示只读模式
+        with h5py.File(file_path, 'r') as f:
+            print("📂 数据集的key列表:")
+
+            # ✅ 递归函数，仅收集数据集的key
+            def collect_dataset_keys(name, obj):
+                if isinstance(obj, h5py.Dataset):
+                    print(f"📊 {name}")
+
+            # ✅ 遍历文件结构，打印数据集key
+            f.visititems(collect_dataset_keys)
+
+            if not any(isinstance(obj, h5py.Dataset) for _, obj in f.items()):
+                print("⚠️ 文件中没有数据集！")
+
+    except FileNotFoundError:
+        print("❌ 文件不存在，请检查文件路径！")
+    except Exception as e:
+        print(f"❌ 发生错误: {str(e)}")
+
+
+def print_dataset_content(file_path, dataset_key):
+    """
+    打印HDF5文件中指定数据集key的内容
+    :param file_path: HDF5文件路径
+    :param dataset_key: 数据集的key（名称路径，如 'demo_1/data'）
+    """
+    try:
+        # ✅ 打开HDF5文件，'r'表示只读模式
+        with h5py.File(file_path, 'r') as f:
+            # ✅ 检查指定key是否存在
+            if dataset_key not in f:
+                print(f"❌ 数据集 '{dataset_key}' 不存在！")
+                return
+
+            # ✅ 获取数据集
+            dataset = f[dataset_key]
+            if not isinstance(dataset, h5py.Dataset):
+                print(f"❌ '{dataset_key}' 不是一个数据集！")
+                return
+
+            # ✅ 打印数据集信息
+            print(f"📊 指定key对应的数据集: {dataset_key}")
+            print(f"  Shape: {dataset.shape} ✅")
+            print(f"  Dtype: {dataset.dtype} ✅")
+
+            # ✅ 打印数据内容
+            if dataset.size < 10:
+                print(f"  Data: {dataset[()]} 📈")
+            else:
+                print(f"  Data (first 1 elements): {dataset[:1]} 📈")
+
+    except FileNotFoundError:
+        print("❌ 文件不存在，请检查文件路径！")
+    except Exception as e:
+        print(f"❌ 发生错误: {str(e)}")
+
+# # 示例用法
+# if __name__ == "__main__":
+#     # 使用你提供的HDF5文件路径
+#     file_path = "/home/sythoid_01/文档/Huangwenlong/LIBERO/libero/datasets/libero_object/pick_up_the_alphabet_soup_and_place_it_in_the_basket_demo.hdf5"
+#     # 替换为你想查看的数据集key
+#     dataset_key = "data/demo_0/states"  # 示例key，需替换为实际的key
+#     print_dataset_content(file_path, dataset_key)
+
+# # 示例用法
+# if __name__ == "__main__":
+#     # 使用你提供的HDF5文件路径
+#     file_path = "/home/sythoid_01/文档/Huangwenlong/LIBERO/libero/datasets/libero_object/pick_up_the_alphabet_soup_and_place_it_in_the_basket_demo.hdf5"
+#     print_dataset_keys(file_path)
+
+# 示例用法
+if __name__ == "__main__":
+    # 替换为你自己的HDF5文件路径
+    file_path = "/home/sythoid_01/文档/Huangwenlong/LIBERO/libero/datasets/libero_object/pick_up_the_alphabet_soup_and_place_it_in_the_basket_demo.hdf5"
+    explore_hdf5(file_path)

custom/demo_array_convert_to_video_test.py

@@ -0,0 +1,80 @@
+import h5py
+import numpy as np
+import cv2
+
+def create_side_by_side_video(file_path, output_video_path, agent_key="data/demo_0/obs/agentview_rgb", eye_key="data/demo_0/obs/eye_in_hand_rgb"):
+    """
+    从HDF5文件中提取agentview_rgb和eye_in_hand_rgbd，生成左右拼接的视频并保存
+    :param file_path: HDF5文件路径
+    :param output_video_path: 输出视频路径（.mp4）
+    :param agent_key: agentview_rgb数据集的key
+    :param eye_key: eye_in_hand_rgbd数据集的key
+    """
+    try:
+        # ✅ 打开HDF5文件
+        with h5py.File(file_path, 'r') as f:
+            # ✅ 检查数据集是否存在
+            if agent_key not in f or eye_key not in f:
+                print(f"❌ 数据集 '{agent_key}' 或 '{eye_key}' 不存在！")
+                return
+
+            # ✅ 读取数据集
+            agent_data = f[agent_key][()]
+            eye_data = f[eye_key][()]
+
+            # ✅ 检查数据形状
+            if agent_data.shape[0] != eye_data.shape[0]:
+                print(f"❌ 两数据集帧数不一致！agentview_rgb: {agent_data.shape[0]}, eye_in_hand_rgbd: {eye_data.shape[0]}")
+                return
+
+            # ✅ 处理RGB和RGBD数据
+            if len(agent_data.shape) != 4 or agent_data.shape[3] != 3:
+                print(f"❌ agentview_rgb 数据形状异常: {agent_data.shape}，期望 (T, H, W, 3)")
+                return
+
+            # 如果eye_in_hand_rgbd是RGBD（4通道），提取RGB部分
+            if len(eye_data.shape) == 4 and eye_data.shape[3] == 4:
+                print("📌 检测到eye_in_hand_rgbd为RGBD，提取RGB通道")
+                eye_data = eye_data[:, :, :, :3]  # 取前3个通道（RGB）
+            elif len(eye_data.shape) != 4 or eye_data.shape[3] != 3:
+                print(f"❌ eye_in_hand_rgbd 数据形状异常: {eye_data.shape}，期望 (T, H, W, 3) 或 (T, H, W, 4)")
+                return
+
+            # ✅ 确保数据类型为uint8
+            if agent_data.dtype != np.uint8:
+                agent_data = agent_data.astype(np.uint8)
+            if eye_data.dtype != np.uint8:
+                eye_data = eye_data.astype(np.uint8)
+
+            # ✅ 获取视频参数
+            num_frames, height, width, _ = agent_data.shape
+            fps = 30  # 可根据需要调整帧率
+            output_size = (width * 2, height)  # 左右拼接，宽度翻倍
+
+            # ✅ 初始化视频写入器
+            fourcc = cv2.VideoWriter_fourcc(*'mp4v')  # 使用mp4v编码
+            out = cv2.VideoWriter(output_video_path, fourcc, fps, output_size)
+
+            # ✅ 逐帧拼接并写入视频
+            for i in range(num_frames):
+                # 左右拼接图像
+                combined_frame = np.hstack((agent_data[i], eye_data[i]))
+                # OpenCV使用BGR格式，需转换RGB到BGR
+                combined_frame = cv2.cvtColor(combined_frame, cv2.COLOR_RGB2BGR)
+                out.write(combined_frame)
+                print(f"📽️ 已处理帧 {i+1}/{num_frames}")
+
+            # ✅ 释放视频写入器
+            out.release()
+            print(f"🎉 视频已保存到: {output_video_path}")
+
+    except FileNotFoundError:
+        print("❌ 文件不存在，请检查文件路径！")
+    except Exception as e:
+        print(f"❌ 发生错误: {str(e)}")
+
+# 示例用法
+if __name__ == "__main__":
+    file_path = "/home/sythoid_01/文档/Huangwenlong/LIBERO/libero/datasets/libero_object/pick_up_the_alphabet_soup_and_place_it_in_the_basket_demo.hdf5"
+    output_video_path = "/home/sythoid_01/文档/Huangwenlong/LIBERO/custom/output_video.mp4"  # 输出视频文件名
+    create_side_by_side_video(file_path, output_video_path)