Unity:Development Buildと通常Buildでの動作の違い

UnityのPC向けアプリを制作しており、
内容は大まかに次の3つのシーンをSceneManagerで切り替えて使用しています。

シーンA:タイトル画面
シーンB:Barracudaのニューラルネットワークを用いたGPU、メモリ共に処理の高いシーン
シーンC:Webカメラを表示するだけの処理の軽いシーン

エディタ上のプレイとDevelopment Buildをしたアプリでは
どのシーンの遷移でもエラーなく動作するのですが、
Development Buildのチェックを外して通常Buildしたアプリでは
シーンBからシーンA、シーンCに遷移する際にその遷移先のシーンの読み込み中にクラッシュを起こして終了しました。

メモリ負荷の落差が原因かと思い、遷移前にシーンBのオブジェクトをすべて消してから遷移するようにしてみましたが解決しませんでした。

Development Buildかそうでないかはアプリ上にエラーログを表示したり、エディターと繋いでログを確認できたりするだけで
処理の動作に違いはないと思っていたのですが他にも違いがあるのでしょうか？

またDevelopment Buildをした場合でもアプリ上にエラーログ等を表示しないようにできないでしょうか？

知恵をお貸しください、よろしくお願いいたします。

なお使っているニューラルネットワークは映像から人の姿勢を推定するものです。

C#
1using UnityEngine;
2using UnityEngine.UI;
3using System.Collections;
4using System.Collections.Generic;
5using Unity.Barracuda;
6
7/// <summary>
8/// Define Joint points
9/// </summary>
10public class Barracuda : MonoBehaviour
11{
12    /// <summary>
13    /// Neural network model
14    /// </summary>
15    public NNModel NNModel;
16
17    public WorkerFactory.Type WorkerType = WorkerFactory.Type.Auto;
18    public bool Verbose = true;
19
20    public VNectModel VNectModel;
21
22    public VideoCapture videoCapture;
23
24    private Model _model;
25    private IWorker _worker;
26
27    /// <summary>
28    /// Coordinates of joint points
29    /// </summary>
30    private VNectModel.JointPoint[] jointPoints;
31    
32    /// <summary>
33    /// Number of joint points
34    /// </summary>
35    private const int JointNum = 24;
36
37    /// <summary>
38    /// input image size
39    /// </summary>
40    public int InputImageSize;
41
42    /// <summary>
43    /// input image size (half)
44    /// </summary>
45    private float InputImageSizeHalf;
46
47    /// <summary>
48    /// column number of heatmap
49    /// </summary>
50    public int HeatMapCol;
51    private float InputImageSizeF;
52
53    /// <summary>
54    /// Column number of heatmap in 2D image
55    /// </summary>
56    private int HeatMapCol_Squared;
57    
58    /// <summary>
59    /// Column nuber of heatmap in 3D model
60    /// </summary>
61    private int HeatMapCol_Cube;
62    private float ImageScale;
63
64    /// <summary>
65    /// Buffer memory has 2D heat map
66    /// </summary>
67    private float[] heatMap2D;
68
69    /// <summary>
70    /// Buffer memory has offset 2D
71    /// </summary>
72    private float[] offset2D;
73    
74    /// <summary>
75    /// Buffer memory has 3D heat map
76    /// </summary>
77    private float[] heatMap3D;
78    
79    /// <summary>
80    /// Buffer memory hash 3D offset
81    /// </summary>
82    private float[] offset3D;
83    private float unit;
84    
85    /// <summary>
86    /// Number of joints in 2D image
87    /// </summary>
88    private int JointNum_Squared = JointNum * 2;
89    
90    /// <summary>
91    /// Number of joints in 3D model
92    /// </summary>
93    private int JointNum_Cube = JointNum * 3;
94
95    /// <summary>
96    /// HeatMapCol * JointNum
97    /// </summary>
98    private int HeatMapCol_JointNum;
99
100    /// <summary>
101    /// HeatMapCol * JointNum_Squared
102    /// </summary>
103    private int CubeOffsetLinear;
104
105    /// <summary>
106    /// HeatMapCol * JointNum_Cube
107    /// </summary>
108    private int CubeOffsetSquared;
109
110    /// <summary>
111    /// For Kalman filter parameter Q
112    /// </summary>
113    public float KalmanParamQ;
114
115    /// <summary>
116    /// For Kalman filter parameter R
117    /// </summary>
118    public float KalmanParamR;
119
120    /// <summary>
121    /// Lock to update VNectModel
122    /// </summary>
123    private bool Lock = true;
124
125    /// <summary>
126    /// Use low pass filter flag
127    /// </summary>
128    public bool UseLowPassFilter;
129
130    /// <summary>
131    /// For low pass filter
132    /// </summary>
133    public float LowPassParam;
134
135    public Text Msg;
136    public float WaitTimeModelLoad = 10f;
137    private float Countdown = 0;
138    public Texture2D InitImg;
139
140    private void Start()
141    {
142        // Initialize 
143        HeatMapCol_Squared = HeatMapCol * HeatMapCol;
144        HeatMapCol_Cube = HeatMapCol * HeatMapCol * HeatMapCol;
145        HeatMapCol_JointNum = HeatMapCol * JointNum;
146        CubeOffsetLinear = HeatMapCol * JointNum_Cube;
147        CubeOffsetSquared = HeatMapCol_Squared * JointNum_Cube;
148
149        heatMap2D = new float[JointNum * HeatMapCol_Squared];
150        offset2D = new float[JointNum * HeatMapCol_Squared * 2];
151        heatMap3D = new float[JointNum * HeatMapCol_Cube];
152        offset3D = new float[JointNum * HeatMapCol_Cube * 3];
153        unit = 1f / (float)HeatMapCol;
154        InputImageSizeF = InputImageSize;
155        InputImageSizeHalf = InputImageSizeF / 2f;
156        ImageScale = InputImageSize / (float)HeatMapCol;// 224f / (float)InputImageSize;
157
158        // Disabel sleep
159        Screen.sleepTimeout = SleepTimeout.NeverSleep;
160
161        // Init model
162        _model = ModelLoader.Load(NNModel, Verbose);
163        _worker = WorkerFactory.CreateWorker(WorkerType, _model, Verbose);
164
165        StartCoroutine("WaitLoad");
166
167    }
168
169    private void Update()
170    {
171        if (!Lock)
172        {
173            UpdateVNectModel();
174        }
175    }
176
177    private IEnumerator WaitLoad()
178    {
179        inputs[inputName_1] = new Tensor(InitImg);
180        inputs[inputName_2] = new Tensor(InitImg);
181        inputs[inputName_3] = new Tensor(InitImg);
182
183        // Create input and Execute model
184        yield return _worker.StartManualSchedule(inputs);
185
186        // Get outputs
187        for (var i = 2; i < _model.outputs.Count; i++)
188        {
189            b_outputs[i] = _worker.PeekOutput(_model.outputs[i]);
190        }
191
192        // Get data from outputs
193        offset3D = b_outputs[2].data.Download(b_outputs[2].shape);
194        heatMap3D = b_outputs[3].data.Download(b_outputs[3].shape);
195
196        // Release outputs
197        for (var i = 2; i < b_outputs.Length; i++)
198        {
199            b_outputs[i].Dispose();
200        }
201
202        // Init VNect model
203        jointPoints = VNectModel.Init();
204
205        PredictPose();
206
207        yield return new WaitForSeconds(WaitTimeModelLoad);
208
209        // Init VideoCapture
210        videoCapture.Init(InputImageSize, InputImageSize);
211        Lock = false;
212        Msg.gameObject.SetActive(false);
213    }
214
215    private const string inputName_1 = "input.1";
216    private const string inputName_2 = "input.4";
217    private const string inputName_3 = "input.7";
218    /*
219    private const string inputName_1 = "0";
220    private const string inputName_2 = "1";
221    private const string inputName_3 = "2";
222    */
223
224    private void UpdateVNectModel()
225    {
226        input = new Tensor(videoCapture.MainTexture);
227        if (inputs[inputName_1] == null)
228        {
229            inputs[inputName_1] = input;
230            inputs[inputName_2] = new Tensor(videoCapture.MainTexture);
231            inputs[inputName_3] = new Tensor(videoCapture.MainTexture);
232        }
233        else
234        {
235            inputs[inputName_3].Dispose();
236
237            inputs[inputName_3] = inputs[inputName_2];
238            inputs[inputName_2] = inputs[inputName_1];
239            inputs[inputName_1] = input;
240        }
241
242        StartCoroutine(ExecuteModelAsync());
243    }
244
245    /// <summary>
246    /// Tensor has input image
247    /// </summary>
248    /// <returns></returns>
249    Tensor input = new Tensor();
250    Dictionary<string, Tensor> inputs = new Dictionary<string, Tensor>() { { inputName_1, null }, { inputName_2, null }, { inputName_3, null }, };
251    Tensor[] b_outputs = new Tensor[4];
252
253    private IEnumerator ExecuteModelAsync()
254    {
255        // Create input and Execute model
256        yield return _worker.StartManualSchedule(inputs);
257
258        // Get outputs
259        for (var i = 2; i < _model.outputs.Count; i++)
260        {
261            b_outputs[i] = _worker.PeekOutput(_model.outputs[i]);
262        }
263
264        // Get data from outputs
265        offset3D = b_outputs[2].data.Download(b_outputs[2].shape);
266        heatMap3D = b_outputs[3].data.Download(b_outputs[3].shape);
267        
268        // Release outputs
269        for (var i = 2; i < b_outputs.Length; i++)
270        {
271            b_outputs[i].Dispose();
272        }
273
274        PredictPose();
275    }
276
277    /// <summary>
278    /// Predict positions of each of joints based on network
279    /// </summary>
280    private void PredictPose()
281    {
282　　　以下略
283　　}
284
285}