deep sort : Simple Online and Realtime Tracking with a Deep Association Metric

ref:

• deep sort
• Deep SORT多目標跟蹤算法代碼解析

follow 他的 README，clong 下來。
依照 instruction run deep_sort_app.py 的話，需要 download Multi-Object-Tracking 數據集。
其中MOT16 的 sequence data，就在最下面..寫 "Download" 的地方..
Download 下來，解開到 MOT16 目錄下。

README 說明:" download pre-generated detections and the CNN checkpoint file from here."
下載 , 是 detections 目錄，放到 resources 目錄下。

不知道不是 python 版本的問題，tunple 的標示操作Error。
所以參考 這一篇，修改 linear_assignment.py:

diff --git a/deep_sort/linear_assignment.py b/deep_sort/linear_assignment.py
index 178456c..d24ec44 100644
--- a/deep_sort/linear_assignment.py
+++ b/deep_sort/linear_assignment.py
@@ -1,7 +1,7 @@
 # vim: expandtab:ts=4:sw=4
 from __future__ import absolute_import
 import numpy as np
-from sklearn.utils.linear_assignment_ import linear_assignment
+from scipy.optimize import linear_sum_assignment as linear_assignment
 from . import kalman_filter


@@ -55,16 +55,16 @@ def min_cost_matching(
     cost_matrix = distance_metric(
         tracks, detections, track_indices, detection_indices)
     cost_matrix[cost_matrix > max_distance] = max_distance + 1e-5
-    indices = linear_assignment(cost_matrix)
+    row_indices, col_indices = linear_assignment(cost_matrix)

     matches, unmatched_tracks, unmatched_detections = [], [], []
     for col, detection_idx in enumerate(detection_indices):
-        if col not in indices[:, 1]:
+        if col not in col_indices:
             unmatched_detections.append(detection_idx)
     for row, track_idx in enumerate(track_indices):
-        if row not in indices[:, 0]:
+        if row not in row_indices:
             unmatched_tracks.append(track_idx)
-    for row, col in indices:
+    for row, col in zip(row_indices,col_indices):
         track_idx = track_indices[row]
         detection_idx = detection_indices[col]
         if cost_matrix[row, col] > max_distance:

最前面是 linear_assignment package 位置改變。
還會有 numpy int 的問題，最後用 python3.5成功（因為 conda 的 python3 最低只有到 3.5)
-- 其實是numpy 版本問題

然後依照說明 run:

python deep_sort_app.py \
    --sequence_dir=./MOT16/test/MOT16-06 \
    --detection_file=./resources/detections/MOT16_POI_test/MOT16-06.npy \
    --min_confidence=0.3 \
    --nn_budget=100 \
    --display=True

就可以看到 tracking 的結果。

之後， python 3.9.16, numpy 1.23.5 也可以順利執行修改過的 code。

README 後段，重新 detection，要使用 tensorflow。
原來應該是使用 tensorflow 1.X
改用 tensorflow 2.4.1 ，要修改：

diff --git a/tools/freeze_model.py b/tools/freeze_model.py
index e89ad29..63b9fc5 100644
--- a/tools/freeze_model.py
+++ b/tools/freeze_model.py
@@ -1,7 +1,7 @@
 # vim: expandtab:ts=4:sw=4
 import argparse
-import tensorflow as tf
-import tensorflow.contrib.slim as slim
+import tensorflow.compat.v1 as tf
+import tf_slim as slim


 def _batch_norm_fn(x, scope=None):
@@ -193,6 +193,7 @@ def parse_args():

 def main():
     args = parse_args()
+    tf.disable_v2_behavior()

     with tf.Session(graph=tf.Graph()) as session:
         input_var = tf.placeholder(

之後，要 freeze model，除了要改 compat.v1，還要修改slim：

diff --git a/tools/freeze_model.py b/tools/freeze_model.py
index e89ad29..63b9fc5 100644
--- a/tools/freeze_model.py
+++ b/tools/freeze_model.py
@@ -1,7 +1,7 @@
 # vim: expandtab:ts=4:sw=4
 import argparse
-import tensorflow as tf
-import tensorflow.contrib.slim as slim
+import tensorflow.compat.v1 as tf
+import tf_slim as slim


 def _batch_norm_fn(x, scope=None):
@@ -193,6 +193,7 @@ def parse_args():

 def main():
     args = parse_args()
+    tf.disable_v2_behavior()

     with tf.Session(graph=tf.Graph()) as session:
         input_var = tf.placeholder(