Upload libero_plus/visualize_hdf.py

libero_plus/visualize_hdf.py

@@ -0,0 +1,162 @@
+import os
+import h5py
+import json
+import numpy as np
+import matplotlib.pyplot as plt
+from matplotlib.patches import Patch
+from pathlib import Path
+
+def extract_task_info_from_filename(filename):
+    """从 HDF5 文件名提取 task_name 和 task_description"""
+    stem = Path(filename).stem  # e.g., "KITCHEN_SCENE3_..."
+    task_name = stem
+    if stem.endswith("_demo"):
+        stem = stem[:-5]
+
+    parts = stem.split('_')
+    # 跳过场景前缀：KITCHEN, SCENE3, LIVING_ROOM, BEDROOM 等
+    i = 0
+    while i < len(parts):
+        part = parts[i]
+        if part.upper() in ["KITCHEN", "LIVING", "BEDROOM", "DINING", "ROOM", "STUDY"] or "SCENE" in part.upper():
+            i += 1
+        else:
+            break
+    description_words = parts[i:]
+    task_description = " ".join(description_words)
+    return task_name, task_description
+
+def visualize_and_save_step(
+    demo_name,
+    agentview_rgb,
+    agentview_seg,
+    obj_name_to_seg_id,
+    save_dir
+):
+    """可视化单个 step 的 RGB + Seg，并保存图像"""
+    rgb_img = np.flipud(agentview_rgb)
+    seg_img = np.flipud(agentview_seg.squeeze())
+
+    fig, axes = plt.subplots(1, 2, figsize=(16, 8))
+
+    # RGB
+    axes[0].imshow(rgb_img)
+    axes[0].set_title("AgentView RGB", fontsize=14)
+    axes[0].axis("off")
+
+    # Segmentation
+    unique_ids = np.unique(seg_img)
+    cmap = plt.get_cmap("tab20", len(unique_ids))
+    im = axes[1].imshow(seg_img, cmap=cmap, vmin=unique_ids.min(), vmax=unique_ids.max())
+    axes[1].set_title("Segmentation", fontsize=14)
+    axes[1].axis("off")
+
+    # 图例
+    seg_id_to_name = {v: k for k, v in obj_name_to_seg_id.items()}
+    legend_elements = []
+    
+    # 获取全局 min/max（用于归一化颜色映射）
+    id_min = unique_ids.min()
+    id_max = unique_ids.max()
+
+    for seg_id in sorted(unique_ids):
+        name = seg_id_to_name.get(seg_id, f"unknown_{int(seg_id)}")
+        # ✅ 正确使用 .min() 和 .max()
+        normed_color = (seg_id - id_min) / (id_max - id_min) if id_max > id_min else 0.0
+        color = cmap(normed_color)
+        legend_elements.append(Patch(facecolor=color, label=f"ID {int(seg_id)}: {name}"))
+
+    plt.legend(handles=legend_elements, bbox_to_anchor=(1.05, 1), loc='upper left', fontsize=10)
+    plt.tight_layout()
+
+    output_path = os.path.join(save_dir, f"{demo_name}_step_20_visualization.png")
+    plt.savefig(output_path, dpi=150, bbox_inches="tight")
+    plt.close(fig)
+    print(f"✅ Saved visualization: {output_path}")
+
+def process_all_hdf5_files(data_dir="."):
+    """处理所有 HDF5 文件，每个文件最多可视化 5 个 demo 的 step 20"""
+    hdf5_files = [f for f in os.listdir(data_dir) if f.endswith(".hdf5")]
+    if not hdf5_files:
+        print("⚠️ No .hdf5 files found.")
+        return
+
+    for file_name in hdf5_files:
+        file_path = os.path.join(data_dir, file_name)
+        try:
+            print(f"\n🔍 Processing: {file_name}")
+            task_name, task_description = extract_task_info_from_filename(file_name)
+            viz_dir = os.path.join("task_visualizations/libero_spatial", task_name)
+            os.makedirs(viz_dir, exist_ok=True)
+
+            desc_file_path = os.path.join(viz_dir, "task_description_id.txt")
+            obj_name_to_seg_id = None
+
+            with h5py.File(file_path, "r") as f:
+                data_group = f["data"]
+                demos = list(data_group.keys())
+                if not demos:
+                    print(f"❌ No demos in {file_name}")
+                    continue
+
+                # 先找一个 demo 来提取 obj_name_to_seg_id（用于写文件）
+                sample_demo = None
+                for demo_name in sorted(demos):
+                    demo_grp = data_group[demo_name]
+                    if 'obj_name_to_seg_id' in demo_grp.attrs:
+                        raw_data = demo_grp.attrs['obj_name_to_seg_id']
+                        if isinstance(raw_data, bytes):
+                            raw_data = raw_data.decode('utf-8')
+                        obj_name_to_seg_id = json.loads(raw_data)
+                        sample_demo = demo_grp
+                        break
+
+                if obj_name_to_seg_id is None:
+                    print(f"❌ Could not find 'obj_name_to_seg_id' in any demo of {file_name}")
+                    continue
+
+                # ✅ 现在可以安全写入文件了
+                if not os.path.exists(desc_file_path):
+                    with open(desc_file_path, "w") as f:
+                        f.write(f"{obj_name_to_seg_id}\n{task_description.strip()}\n{task_description.strip()}")
+                    print(f"📝 Saved mapping and description to: {desc_file_path}")
+
+                # 选择最多 5 个 demo 进行可视化
+                selected_demos = sorted(demos)[:5]
+
+                for demo_name in selected_demos:
+                    demo_grp = data_group[demo_name]
+                    obs = demo_grp["obs"]
+
+                    # 检查长度
+                    num_steps = obs["agentview_rgb"].shape[0]
+                    if num_steps <= 20:
+                        print(f"🟡 Skipping {demo_name}: only {num_steps} steps (<21)")
+                        continue
+
+                    agentview_rgb = obs["agentview_rgb"][20]
+                    agentview_seg = obs["agentview_seg"][20]
+
+                    # 获取当前 demo 的 obj_name_to_seg_id（确保一致性）
+                    if 'obj_name_to_seg_id' in demo_grp.attrs:
+                        raw_data = demo_grp.attrs['obj_name_to_seg_id']
+                        if isinstance(raw_data, bytes):
+                            raw_data = raw_data.decode('utf-8')
+                        current_mapping = json.loads(raw_data)
+                    else:
+                        current_mapping = obj_name_to_seg_id  # fallback
+
+                    # 可视化
+                    visualize_and_save_step(
+                        demo_name=demo_name,
+                        agentview_rgb=agentview_rgb,
+                        agentview_seg=agentview_seg,
+                        obj_name_to_seg_id=current_mapping,
+                        save_dir=viz_dir
+                    )
+
+        except Exception as e:
+            print(f"❌ Error processing {file_name}: {e}")
+
+if __name__ == "__main__":
+    process_all_hdf5_files("/mnt/data/data/libero_plus/libero_spatial")