Refactor Open3D Visualizer and Add Utilities for Method Comparison (#28)

* chore(visualization): refactor the open3d visualization, merge fn together.
* Reset fire from function to class usage directly.
* add save screenshot easily with multi-view.
* sync viewpoint through diff windows.
* visual lidar center tf if set slc to True.
* fix(flow): add index_flow for 2hz gt view etc.

---------

Co-authored-by: lisiyi777 <lisiyi777@users.noreply.github.com>

Author: Kin-Zhang

README.md

@@ -301,7 +301,7 @@ pip install "evalai"
 evalai set-token <your token>
 
 # Step 3: Copy the command pop above and submit to leaderboard
-evalai challenge 2010 phase 4018 submit --file av2_submit.zip --large --private
+# evalai challenge 2010 phase 4018 submit --file av2_submit.zip --large --private
 evalai challenge 2210 phase 4396 submit --file av2_submit_v2.zip --large --private
 ```
 
@@ -318,29 +318,31 @@ python eval.py model=nsfp dataset_path=/home/kin/data/av2/h5py/demo/val
 # The output of above command will be like:
 Model: DeFlow, Checkpoint from: /home/kin/model_zoo/v2/seflow_best.ckpt
 We already write the flow_est into the dataset, please run following commend to visualize the flow. Copy and paste it to your terminal:
-python tools/visualization.py --res_name 'seflow_best' --data_dir /home/kin/data/av2/preprocess_v2/sensor/vis
+python tools/visualization.py vis --res_name 'seflow_best' --data_dir /home/kin/data/av2/preprocess_v2/sensor/vis
 Enjoy! ^v^ ------ 
 
 # Then run the command in the terminal:
-python tools/visualization.py --res_name 'seflow_best' --data_dir /home/kin/data/av2/preprocess_v2/sensor/vis
+python tools/visualization.py vis --res_name 'seflow_best' --data_dir /home/kin/data/av2/preprocess_v2/sensor/vis
 ```
 
 https://github.com/user-attachments/assets/f031d1a2-2d2f-4947-a01f-834ed1c146e6
 
 For exporting easy comparsion with ground truth and other methods, we also provided multi-visulization open3d window:
 ```bash
-python tools/visualization.py --mode mul --res_name "['flow', 'seflow_best']" --data_dir /home/kin/data/av2/preprocess_v2/sensor/vis
+python tools/visualization.py vis --res_name "['flow', 'seflow_best']" --data_dir /home/kin/data/av2/preprocess_v2/sensor/vis
 ```
 
-Or another way to interact with [rerun](https://github.com/rerun-io/rerun) but please only vis scene by scene, not all at once.
+**Tips**: To quickly create qualitative results for all methods, you can use multiple results comparison mode, select a good viewpoint and then save screenshots for all frames by pressing `P` key. You will found all methods' results are saved in the output folder (default is `logs/imgs`). Enjoy it!
+
+
+_Rerun_: Another way to interact with [rerun](https://github.com/rerun-io/rerun) but please only vis scene by scene, not all at once.
 
 ```bash
 python tools/visualization_rerun.py --data_dir /home/kin/data/av2/h5py/demo/train --res_name "['flow', 'deflow']"
 ```
 
 https://github.com/user-attachments/assets/07e8d430-a867-42b7-900a-11755949de21
 
-
 ## Cite Us
 
 [*OpenSceneFlow*](https://github.com/KTH-RPL/OpenSceneFlow) is originally designed by [Qingwen Zhang](https://kin-zhang.github.io/) from DeFlow and SeFlow. 

src/dataset.py

@@ -187,7 +187,7 @@ class HDF5Dataset(Dataset):
     def __init__(self, directory, \
                 transform=None, n_frames=2, ssl_label=None, \
                 eval = False, leaderboard_version=1, \
-                vis_name=''):
+                vis_name='', index_flow=False):
         '''
         Args:
             directory: the directory of the dataset, the folder should contain some .h5 file and index_total.pkl.
@@ -199,6 +199,7 @@ def __init__(self, directory, \
             * eval: if True, use the eval index (only used it for leaderboard evaluation)
             * leaderboard_version: 1st or 2nd, default is 1. If '2', we will use the index_eval_v2.pkl from assets/docs.
             * vis_name: the data of the visualization, default is ''.
+            * index_flow: if True, use the flow index for training or visualization.
         '''
         super(HDF5Dataset, self).__init__()
         self.directory = directory
@@ -247,7 +248,7 @@ def __init__(self, directory, \
 
         # for some dataset that annotated HZ is different.... like truckscene and nuscene etc.
         self.train_index = None
-        if not eval and ssl_label is None and transform is not None: # transform indicates whether we are in training mode.
+        if (not eval and ssl_label is None and transform is not None) or index_flow: # transform indicates whether we are in training mode.
             # check if train seq all have gt.
             one_scene_id = list(self.scene_id_bounds.keys())[0]
             check_flow_exist = True
@@ -347,7 +348,7 @@ def __getitem__(self, index_):
                     data_dict[f'gmh{i+1}'] = past_gm
                     data_dict[f'poseh{i+1}'] = past_pose
 
-            for data_key in self.vis_name + ['ego_motion', 'lidar_dt', 
+            for data_key in self.vis_name + ['ego_motion', 'lidar_dt', 'lidar_center',
                              # ground truth information:
                              'flow', 'flow_is_valid', 'flow_category_indices', 'flow_instance_id', 'dufo']:
                 if data_key in f[key]:

src/trainer.py

@@ -243,7 +243,7 @@ def on_validation_epoch_end(self):
             with open(str(self.save_res_path)+'.pkl', 'wb') as f:
                 pickle.dump((self.metrics.epe_3way, self.metrics.bucketed, self.metrics.epe_ssf), f)
             print(f"We already write the {self.res_name} into the dataset, please run following commend to visualize the flow. Copy and paste it to your terminal:")
-            print(f"python tools/visualization.py --res_name '{self.res_name}' --data_dir {self.dataset_path}")
+            print(f"python tools/visualization.py vis --res_name '{self.res_name}' --data_dir {self.dataset_path}")
             print(f"Enjoy! ^v^ ------ \n")
 
         self.metrics = OfficialMetrics()

src/utils/mics.py

@@ -172,6 +172,78 @@ def make_colorwheel(transitions: tuple=DEFAULT_TRANSITIONS) -> np.ndarray:
     return colorwheel
 
 
+def error_to_color(
+    error_magnitude: np.ndarray,
+    max_error: float,
+    color_map: str = "jet"
+) -> np.ndarray:
+    """
+    Convert flow error to RGB color visualization.
+    Args:
+        color_map: Color map to use for visualization ("jet" recommended for error visualization)
+        
+    Returns:
+        RGB color representation of the error of shape (..., 3)
+    """
+    if max_error > 0:
+        normalized_error = np.clip(error_magnitude / max_error, 0, 1)
+    else:
+        normalized_error = np.zeros_like(error_magnitude)
+    
+    # Create colormap
+    if color_map == "jet":
+        # Simple jet colormap implementation
+        colors = np.zeros((*normalized_error.shape, 3), dtype=np.uint8)
+        
+        # Blue to cyan to green to yellow to red
+        # Blue (low error)
+        idx = normalized_error < 0.25
+        colors[idx, 2] = 255
+        colors[idx, 0] = 0
+        colors[idx, 1] = np.uint8(255 * normalized_error[idx] * 4)
+        
+        # Cyan to green
+        idx = (normalized_error >= 0.25) & (normalized_error < 0.5)
+        colors[idx, 1] = 255
+        colors[idx, 0] = 0
+        colors[idx, 2] = np.uint8(255 * (1 - (normalized_error[idx] - 0.25) * 4))
+        
+        # Green to yellow
+        idx = (normalized_error >= 0.5) & (normalized_error < 0.75)
+        colors[idx, 1] = 255
+        colors[idx, 2] = 0
+        colors[idx, 0] = np.uint8(255 * (normalized_error[idx] - 0.5) * 4)
+        
+        # Yellow to red (high error)
+        idx = normalized_error >= 0.75
+        colors[idx, 0] = 255
+        colors[idx, 2] = 0
+        colors[idx, 1] = np.uint8(255 * (1 - (normalized_error[idx] - 0.75) * 4))
+    
+    elif color_map == "hot":
+        # Hot colormap: black -> red -> yellow -> white
+        colors = np.zeros((*normalized_error.shape, 3), dtype=np.uint8)
+        
+        # Black to red
+        idx = normalized_error < 0.33
+        colors[idx, 0] = np.uint8(255 * normalized_error[idx] * 3)
+        
+        # Red to yellow
+        idx = (normalized_error >= 0.33) & (normalized_error < 0.67)
+        colors[idx, 0] = 255
+        colors[idx, 1] = np.uint8(255 * (normalized_error[idx] - 0.33) * 3)
+        
+        # Yellow to white
+        idx = normalized_error >= 0.67
+        colors[idx, 0] = 255
+        colors[idx, 1] = 255
+        colors[idx, 2] = np.uint8(255 * (normalized_error[idx] - 0.67) * 3)
+    
+    else:
+        raise ValueError(f"Unsupported color map: {color_map}. Use 'jet' or 'hot'.")
+    
+    return colors
+
 def flow_to_rgb(
     flow: np.ndarray,
     flow_max_radius: Optional[float]=None,