Visualization updates (#5)

* Made visualization load lazy
* Updated eval version

---------

Co-authored-by: Ishan Khatri <[email protected]>

docker/Dockerfile

@@ -27,7 +27,7 @@ RUN pip install --no-index --no-cache-dir pytorch3d -f https://dl.fbaipublicfile
 RUN pip install yapf==0.40.1
 RUN pip install pyarrow pyquaternion open3d tensorboard pytorch-lightning==1.9.4 imageio black isort nntime
 RUN pip install mmcv==2.1.0 -f https://download.openmmlab.com/mmcv/dist/cu113/torch1.12.0/index.html
-RUN pip install bucketed-scene-flow-eval==1.0.1
+RUN pip install bucketed-scene-flow-eval==1.0.4
 RUN apt install -y unzip
 
 ENV PYTHONPATH=/project:/bucketed_scene_flow_eval:/:${PYTHONPATH}

launch.sh

@@ -5,6 +5,7 @@ docker run --gpus=all --rm -it \
  --shm-size=16gb \
  -v `pwd`:/project \
  -v /efs:/efs \
+ -v /efs2:/efs2 \
  -v /bigdata:/bigdata \
  -v /tmp/.X11-unix:/tmp/.X11-unix \
  -v /tmp/frame_results:/tmp/frame_results \

visualization/dataset_count_vis.py

@@ -2,12 +2,12 @@
 from loader_utils import load_pickle, load_json
 import numpy as np
 from typing import Dict, List, Any
-from bucketed_scene_flow_eval.datasets.argoverse2.av2_metacategories import METACATAGORIES
+from bucketed_scene_flow_eval.datasets.argoverse2.av2_metacategories import BUCKETED_METACATAGORIES
 import matplotlib.pyplot as plt
 from visualization.figures.plot_lib import *
 
 # Load the saved count data
-raw_class_speed_count_info = load_json('/efs/argoverse2/val_dataset_stats/dataset_count_info.json')
+raw_class_speed_count_info = load_json("/efs/argoverse2/val_dataset_stats/dataset_count_info.json")
 
 
 def merge_classes(class_a: Dict[float, int], class_b: Dict[float, int]):
@@ -22,15 +22,13 @@ def merge_classes(class_a: Dict[float, int], class_b: Dict[float, int]):
 
 
 def collapse_into_meta_classes(
-        count_info: Dict[str, Dict[float,
-                                   int]]) -> Dict[str, Dict[float, int]]:
+    count_info: Dict[str, Dict[float, int]]
+) -> Dict[str, Dict[float, int]]:
     # Create a new dictionary to hold the meta classes
     meta_class_count_info = {}
     # For each meta class, merge the classes that belong to it
-    for meta_class_name, meta_class_entries in METACATAGORIES.items():
-        entries_list = [
-            count_info[category_name] for category_name in meta_class_entries
-        ]
+    for meta_class_name, meta_class_entries in BUCKETED_METACATAGORIES.items():
+        entries_list = [count_info[category_name] for category_name in meta_class_entries]
         # Merge the classes
         merged_class = entries_list[0]
         for class_to_merge in entries_list[1:]:
@@ -49,8 +47,7 @@ def collapse_speed_buckets(entry: Dict[float, int]) -> int:
 meta_class_speed_info = collapse_into_meta_classes(raw_class_speed_count_info)
 meta_class_count_info = {
     meta_class_name: collapse_speed_buckets(meta_class_speed_info)
-    for meta_class_name, meta_class_speed_info in
-    meta_class_speed_info.items()
+    for meta_class_name, meta_class_speed_info in meta_class_speed_info.items()
 }
 
 
@@ -60,39 +57,41 @@ def plot_meta_class_counts(meta_class_count_info: Dict[str, int]):
 
     total_points = sum(meta_class_count_info.values())
 
-    entries = sorted(meta_class_count_info.items(),
-                     key=lambda x: x[1],
-                     reverse=True)
+    entries = sorted(meta_class_count_info.items(), key=lambda x: x[1], reverse=True)
     labels, values = zip(*entries)
+    labels = [label.replace("_", " ") for label in labels]
 
     bars = ax.bar(labels, values, color=color(0, 1))
-    ax.set_xlabel("Meta Class")
-    ax.set_ylabel("Lidar points in semantic class")
-    ax.set_title("Meta Class Counts")
+    # ax.set_xlabel("Meta Class")
+    ax.set_ylabel("Lidar points in metaclass")
+    # ax.set_title("Meta Class Counts")
 
     # Make the X axis labels at 30 Degrees
-    plt.xticks(rotation=30, ha='right')
+    plt.xticks(rotation=30, ha="right")
     # Make the Y axis log scale
-    ax.set_yscale('log')
+    ax.set_yscale("log")
 
     # Annotate bars with percentage
     for bar in bars:
         height = bar.get_height()
         percentage = 100 * height / total_points
-        ax.text(bar.get_x() + bar.get_width() / 2,
-                height,
-                f'{percentage:.2f}%',
-                ha='center',
-                va='bottom')
+        ax.text(
+            bar.get_x() + bar.get_width() / 2,
+            height,
+            f"{percentage:.2f}%",
+            ha="center",
+            va="bottom",
+        )
 
     # fig.savefig("meta_class_counts.pdf", bbox_inches='tight')
     # Save as a png as well
-        
+
     # Remove the top and right axes
-    ax.spines['right'].set_visible(False)
-    ax.spines['top'].set_visible(False)
+    ax.spines["right"].set_visible(False)
+    ax.spines["top"].set_visible(False)
     savefig("validation_results", "meta_class_counts")
 
+set_font(12)
 
 plot_meta_class_counts(meta_class_count_info)
 

visualization/threeway_epe_eval/copy_data.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+
+# Define the source and target directories
+SOURCE_DIR="/efs2/argoverse2_submission_evals"
+TARGET_DIR="/home/k/code/offline_sceneflow_eval/visualization/perf_data/threeway_epe"
+
+# Iterate over each subfolder in the source directory
+for SUBFOLDER in "$SOURCE_DIR"/*; do
+    # Extract the name of the subfolder
+    FOLDER_NAME=$(basename "$SUBFOLDER")
+    # Define the source file path
+    SOURCE_FILE="${SUBFOLDER}/eval_results/per_class_results_35.json"
+    # Define the target file path, appending the subfolder name for uniqueness
+    TARGET_FILE="${TARGET_DIR}/${FOLDER_NAME}_per_class_results_35.json"
+    # Copy the file from the source to the target directory
+    cp "$SOURCE_FILE" "$TARGET_FILE"
+done
+
+echo "Copy completed."

visualization/threeway_epe_eval/run_evals.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+for f in /efs2/argoverse2_submission_evals/*; do
+    echo "Processing $f file..."
+    # Extract the full filename without the path
+    filename=$(basename -- "$f")
+    python ~/code/bucketed_scene_flow_eval/scripts/evals/av2_eval.py /efs/argoverse2/test/ /efs/argoverse2/test_sceneflow_feather "/efs2/argoverse2_submission_evals/$filename/submission" "/efs2/argoverse2_submission_evals/$filename/eval_results" --eval_type threeway_epe --cache_root /tmp/av2_evals --cpu_count 46
+done

visualization/threeway_epe_eval/setup_files.sh

@@ -0,0 +1,10 @@
+#!/bin/bash
+
+for f in /efs/argoverse2/test_*_submission.zip; do
+    echo "Processing $f file..."
+    # Extract the full filename without the path
+    filename=$(basename -- "$f")
+    # Remove the .zip extension to get the stem
+    stem="${filename%.zip}"
+    python ~/code/bucketed_scene_flow_eval/scripts/evals/setup_sparse_user_submission.py "/efs2/argoverse2_submission_evals/$stem/" $f /efs/argoverse2/test_sceneflow_feather
+done