update

cfg/yolov3-spp-21cls.cfg

@@ -633,14 +633,14 @@ activation=leaky
 size=1
 stride=1
 pad=1
-filters=72
+filters=78
 activation=linear
 
 
 [yolo]
 mask = 6,7,8
 anchors=26,16, 19,45, 36,27, 54,32, 34,80, 60,49, 74,71, 96,105, 135,145
-classes=19
+classes=21
 num=9
 jitter=.3
 ignore_thresh = .7
@@ -719,14 +719,14 @@ activation=leaky
 size=1
 stride=1
 pad=1
-filters=72
+filters=78
 activation=linear
 
 
 [yolo]
 mask = 3,4,5
 anchors=26,16, 19,45, 36,27, 54,32, 34,80, 60,49, 74,71, 96,105, 135,145
-classes=19
+classes=21
 num=9
 jitter=.3
 ignore_thresh = .7
@@ -806,14 +806,14 @@ activation=leaky
 size=1
 stride=1
 pad=1
-filters=72
+filters=78
 activation=linear
 
 
 [yolo]
 mask = 0,1,2
 anchors=26,16, 19,45, 36,27, 54,32, 34,80, 60,49, 74,71, 96,105, 135,145
-classes=19
+classes=21
 num=9
 jitter=.3
 ignore_thresh = .7

cfg/yolov3-spp.cfg

@@ -640,7 +640,7 @@ activation=linear
 [yolo]
 mask = 6,7,8
 anchors = 10,13,  16,30,  33,23,  30,61,  62,45,  59,119,  116,90,  156,198,  373,326
-classes=80
+classes=19
 num=9
 jitter=.3
 ignore_thresh = .7
@@ -726,7 +726,7 @@ activation=linear
 [yolo]
 mask = 3,4,5
 anchors = 10,13,  16,30,  33,23,  30,61,  62,45,  59,119,  116,90,  156,198,  373,326
-classes=80
+classes=19
 num=9
 jitter=.3
 ignore_thresh = .7
@@ -813,7 +813,7 @@ activation=linear
 [yolo]
 mask = 0,1,2
 anchors = 10,13,  16,30,  33,23,  30,61,  62,45,  59,119,  116,90,  156,198,  373,326
-classes=80
+classes=19
 num=9
 jitter=.3
 ignore_thresh = .7

data/widok01-11.data

@@ -1,6 +0,0 @@
-classes=18
-train=/home/michall/yolov3/data/widok01-11_train_labels.txt
-valid=/home/michall/yolov3/data/widok01-11_test_labels.txt
-names=/home/michall/yolov3/data/widok01-11.names
-backup=backup/
-eval=coco

data/widok_01_19.data

@@ -1,6 +0,0 @@
-classes=19
-train=/home/tomekb/yolov3/data/widok_01_19/widok_01_19_train_labels.txt
-valid=/home/tomekb/yolov3/data/widok_01_19/widok_01_19_test_labels.txt
-names=/home/tomekb/yolov3/data/widok_01_19/widok_01_19.names
-backup=backup/
-eval=coco

data/widok_01_21/widok_01_21.data

@@ -0,0 +1,6 @@
+classes=21
+train=./data/widok_01_21/widok_01_21_train_labels.txt
+valid=./data/widok_01_21/widok_01_21_test_labels.txt
+names=./data/widok_01_21/widok_01_21.names
+backup=backup/
+eval=coco

data/widok_01_21/widok_01_21.names

@@ -0,0 +1,21 @@
+1. rower
+2. motocykl
+3. osobowy
+4. osobowy pickup
+5. osobowy dostawczy
+6. osobowy van 7-9
+7. dostawczy blaszak / BUS sredni dostawczy
+8. dostawczy zabudowany
+9. dostawczy pickup (w tym pomoc drog.)
+10. dostawczy VAN (osobowy) / autobus maly 10-24 / BUS sredni osobowy
+11. autobus miejski / autobus turystyczny i inny
+12. ciezarowy pow. 3,5t zabudowany / ciezarowy z widoczna przyczepa
+13. ciezarowy pow. 3,5t otwarty (w tym duzy holownik)
+14. ciezarowy pow. 3,5t inny (wanna, gruszka, dzwig itp.)
+15. ciagnik siodlowy z widoczna naczepa / ciagnik siodlowy bez naczepy
+16. inne pojazdy silnikowe / camper / woz strazacki / ciagnik roliczy, koparka, spychacz
+17. przyczepa
+18. BUS-karetka/policja
+19. BUS brygadowka
+20. BUS sredni dostawczy
+21. BUS sredni osobowy

data/widok_01_21/widok_01_21.names.save

@@ -0,0 +1,16 @@
+1. rower
+2. motocykl
+3. osobowy
+4. osobowy pickup
+5. osobowy dostawczy
+6. osobowy van 7-9
+7. dostawczy blaszak / BUS sredni dostawczy
+8. dostawczy zabudowany
+9. dostawczy pickup (w tym pomoc drog.)
+10. dostawczy VAN (osobowy) / autobus maly 10-24 / BUS sredni osobowy
+11. autobus miejski / autobus turystyczny i inny
+12. ciezarowy pow. 3,5t zabudowany / ciezarowy z widoczna przyczepa
+13. ciezarowy pow. 3,5t otwarty / (w tym duzy holownik)
+14. ciezarowy pow. 3,5t inny (wanna, gruszka, dzwig itp.)
+15. ciagnik siodlowy z widoczna naczepa / 
+ciagnik siodlowy bez naczepy

detect.py

@@ -133,7 +133,7 @@ def detect(save_img=False):
                         plot_one_box(xyxy, im0, label=label, color=colors[int(cls)])
 
             # Print time (inference + NMS)
-            print('%sDone. (%.3fs)' % (s, t2 - t1))
+            #print('%sDone. (%.3fs)' % (s, t2 - t1))
 
             # Stream results
             if view_img:
@@ -162,7 +162,7 @@ def detect(save_img=False):
         if platform == 'darwin':  # MacOS
             os.system('open ' + save_path)
 
-    print('Done. (%.3fs)' % (time.time() - t0))
+    #print('Done. (%.3fs)' % (time.time() - t0))
 
 
 if __name__ == '__main__':
@@ -186,7 +186,7 @@ if __name__ == '__main__':
     opt = parser.parse_args()
     opt.cfg = check_file(opt.cfg)  # check file
     opt.names = check_file(opt.names)  # check file
-    print(opt)
+    #print(opt)
 
     with torch.no_grad():
         detect(True)

our_scripts/config.yml

@@ -1,11 +1,11 @@
 train:
-  epochs: 1200
+  epochs: 200
-  batch-size: 3
+  batch-size: 14 #
   cfg: ./cfg/yolov3-spp-19cls.cfg
   data: ./data/widok_01_19.data
   multi-scale: false
-  img-size: '512 1920'
+  img-size: '768 1280'
-  rect: true
+  rect: false
   resume: false
   nosave: false
   notest: false
@@ -16,7 +16,8 @@ train:
   device: 1
   adam: true
   single-cls: false
-  save-every-nth-epoch: 50
+  snapshot-every: 50
+  freeze-layers: true
 
   # inne hiperparametry
   other-hyps:
@@ -43,7 +44,7 @@ experiments:
   dir: ./experiments
 detect:
   source: /home/tomekb/yolov3/data/widok_01_19/widok_01_19_test_labels.txt
-  test-img-size: 1920
+  test-img-size: 1024
   conf-thres: 0.3
   iou-thres: 0.6
   classes:

our_scripts/config_bayes.py

@@ -1,29 +1,7 @@
 import yaml
 
-
-class Args:
-
-    def get_args_string(self) -> str:
-        string = ''
-        for key, value in self.__dict__.items():
-            if not isinstance(value, Configuration.Train.OtherHyps) and value is not None:
-                if key == 'img-size':
-                    string += f' --{key} {value.split(" ")[0]} {value.split(" ")[1]}'
-                elif type(value) == bool:
-                    if value:
-                        string += f" --{key}"
-                    else:
-                        continue
-                elif type(value) in [int, str] and value != '':
-                    string += f' --{key} {value}'
-                else:
-                    raise Exception(f"Cannot parse argument {key} {value}")
-
-        return string
-
-
 class Configuration:
-    class Train(Args):
+    class Train:
 
         class OtherHyps:
             def __init__(self, config_file) -> None:
@@ -36,28 +14,55 @@ class Configuration:
                 if key != 'other-hyps':
                     self.__dict__[key] = value
 
-    class Experiments(Args):
+    class Experiments:
         def __init__(self, config_file) -> None:
             for key, value in config_file['experiments'].items():
                 self.__dict__[key] = value
 
-    class Detect(Args):
+    class Detect:
         def __init__(self, config_file) -> None:
             for key, value in config_file['detect'].items():
                 self.__dict__[key] = value
 
-    class ConfussionMatrix(Args):
+    class ConfussionMatrix:
         def __init__(self, config_file) -> None:
             for key, value in config_file['confussion-matrix'].items():
                 self.__dict__[key] = value
 
-    def __init__(self, config_path='/home/tomekb/yolov3/our_scripts/config_bayes.yml') -> None:
+    class Bayes:
+        def __init__(self, config_file) -> None:
+            for key, value in config_file['bayes'].items():
+                self.__dict__[key] = value
+
+    def __init__(self, config_path='./config_bayes.yml') -> None:
         self.config_path = config_path
         file = yaml.load(open(config_path, 'r'), Loader=yaml.Loader)
         self.train = self.Train(file)
         self.experiments = self.Experiments(file)
         self.detect = self.Detect(file)
         self.confussion_matrix = self.ConfussionMatrix(file)
+        self.bayes = self.Bayes(file)
+
+    def get_bayes_bounds(self) -> list:
+        result = []
+        dicts = {**self.train.__dict__, **self.train.other_hyps.__dict__, **self.detect.__dict__}
+        for key, value in dicts.items():
+            if type(value) not in [None, Configuration.Train.OtherHyps] and type(value) == dict:
+                if value['type'] == 'continuous':  # continous value
+                    val = (value['min'], value['max'])
+                    item = {'name': key, 'type': value['type'], 'domain': val}
+                elif value['type'] == 'discrete' and 'step' in value:
+                    val = tuple(n for n in range(value['min'], value['max'], value['step']))
+                    item = {'name': key, 'type': value['type'], 'domain': val}
+                elif value['type'] == 'discrete':  # discrete values without step
+                    val = tuple(n for n in value['values'])
+                    item = {'name': key, 'type': value['type'], 'domain': val}
+                else:  # unknown type
+                    raise Exception("Invalid type", value['type'])
+
+                result.append(item)
+        return result
+
 
 if __name__ == '__main__':
     config = Configuration()

our_scripts/config_bayes.yml

@@ -1,32 +1,47 @@
+bayes:
+  iterations: 2
 train:
-  epochs: 100
+  epochs:
+    type: discrete
+    values: [10]
   batch-size:
     type: discrete
-    values: 1,2,3,4
+    #values: [128]
-  cfg: ./cfg/yolov3-spp-19cls.cfg
+    min: 1
-  data: ./data/widok_01_19.data
+    max: 5
+    step: 1
+  cfg: ./cfg/yolov3-spp-21cls.cfg
+  data: ./data/widok_01_21.data
   multi-scale:
     type: discrete
-    values: true, false
+    values: [true, false]
-  img-size:
+  img-size-start:
     type: discrete
-    values: (512, 1280),(576, 1280),(640, 1280),(704, 1280),(768, 1280),(832,1280),(960, 1280),(1024, 1280) # trzeba wziąć pod uwagę wszystkie kombinacje  i warunek img-size-min < img-size_max których jest cała masa
+    min: 512
-  rect: false
+    max: 1088
+    step: 64
+  img-size-end:
     type: discrete
-    values: true,false
+    min: 512
+    max: 1088
+    step: 64
+  rect:
+    type: discrete
+    values: [false]
   resume: false
   nosave: false
   notest: false
   evolve: false
   bucket:
   cache-images: false
-  weights: /home/tomekb/yolov3/weights/yolov3-spp-ultralytics.pt
+  weights: ./weights/yolov3-spp-ultralytics.pt
   device: 1
-  adam: true
+  adam:
+    type: discrete
+    values: [true]
   single-cls: false
-  snapshot-every: 50
+  snapshot-every:
   freeze-layers: true
-
   other-hyps:
     giou:
       type: continuous
@@ -52,8 +67,14 @@ train:
       type: continuous
       min: 0.0
       max: 1.0
-    lr0: 0.01  # initial learning rate (SGD=5E-3 Adam=5E-4) # trzeba wziąć pod uwage zależność lr0 < lrf dlatego nie zmieniam
+    lr0:
-    lrf: 0.0005  # final learning rate (with cos scheduler)
+      type: continuous
+      min: 0.000001
+      max: 0.1
+    lrf:
+      type: continuous
+      min: 0.000001
+      max: 0.1
     momentum:
       type: continuous
       min: 0.0
@@ -70,7 +91,7 @@ train:
       type: continuous
       min: 0.0
       max: 1.0
-    hsv_s: 0.678
+    hsv_s:
       type: continuous
       min: 0.0
       max: 1.0
@@ -97,10 +118,12 @@ train:
 experiments:
   dir: ./experiments
 detect:
-  source: /home/tomekb/yolov3/data/widok_01_19/widok_01_19_test_labels.txt
+  source: ./data/widok_01_21/widok_01_21_test_labels.txt
   test-img-size:
     type: discrete
-    values: 512,576,640,704,768,832,896,960,1024,1088,1152,1216,1280,
+    min: 512
+    max: 1088
+    step: 64
   conf-thres:
     type: continuous
     min: 0.0
@@ -113,4 +136,4 @@ detect:
   agnostic-nms:
   augment:
 confussion-matrix:
-  labels-dir: /home/tomekb/yolov3/data/widok_01_19/widok_01_19_labels
+  labels-dir: ./data/widok_01_21/widok_01_21_labels

our_scripts/generate-confusion-matrix.js

@@ -158,10 +158,10 @@ async function main() {
     tsvRows.push([ names[row], ...(summaryRows.map(r => summaryPart[r] && (summaryPart[r] / sum).toFixed(2)))].join('\t'))
   }
 
-  const allLabeled = rows.slice(0, -1).map(r => summary.sum[r]).reduce((a, b) => a + b, 0)
+  const allDetected = rows.slice(0, -1).map(r => summary.sum[r]).reduce((a, b) => a + b, 0)
-  const allDetected = rows.slice(0, -1).map(r => summary[r].sum).reduce((a, b) => a + b, 0)
+  const allLabeled = rows.slice(0, -1).map(r => summary[r].sum).reduce((a, b) => a + b, 0)
-  const falseNegatives = rows.slice(0, -1).map(r => summary.n[r] || 0).reduce((a, b) => a + b, 0)
+  const falsePositives = rows.slice(0, -1).map(r => summary.n[r] || 0).reduce((a, b) => a + b, 0)
-  const falsePositives = rows.slice(0, -1).map(r => summary[r].n || 0).reduce((a, b) => a + b, 0)
+  const falseNegatives = rows.slice(0, -1).map(r => summary[r].n || 0).reduce((a, b) => a + b, 0)
   const right = rows.slice(0, -1).map(r => summary[r][r] || 0).reduce((a, b) => a + b, 0)
   const mistakes = rows.slice(0, -1).map(a => rows.slice(0, -1).map(b => (a!=b && summary[a][b]) || 0).reduce((a, b) => a + b, 0)).reduce((a, b) => a + b, 0)
 

our_scripts/generate_txt_from_xml.py

@@ -46,7 +46,7 @@ def map_class_name_to_id(class_name, xml_document, class_distribution):
     elif class_name in ['6. osobowy van 7-9']:
         class_distribution[5] += 1
         return 5
-    elif class_name in ['7. dostawczy blaszak', '27. BUS sredni dostawczy']:
+    elif class_name in ['7. dostawczy blaszak']:
         class_distribution[6] += 1
         return 6
     elif class_name in ['8. dostawczy zabudowany']:
@@ -55,7 +55,7 @@ def map_class_name_to_id(class_name, xml_document, class_distribution):
     elif class_name in ['9. dostawczy pickup (w tym pomoc drog.)']:
         class_distribution[8] += 1
         return 8
-    elif class_name in ['10. dostawczy VAN (osobowy)', '11. autobus maly 10-24', '28. BUS sredni osobowy']:
+    elif class_name in ['10. dostawczy VAN (osobowy)', '11. autobus maly 10-24']:
         class_distribution[9] += 1
         return 9
     elif class_name in ['12. autobus miejski', '13. autobus turystyczny i inny']:
@@ -85,13 +85,19 @@ def map_class_name_to_id(class_name, xml_document, class_distribution):
     elif class_name in ['26. BUS brygadowka']:
         class_distribution[18] += 1
         return 18
+    elif class_name in ['27. BUS sredni dostawczy']:
+        class_distribution[19] += 1
+        return 19
+    elif class_name in ['28. BUS sredni osobowy']:
+        class_distribution[20] += 1
+        return 20
     else:
         raise Exception('Unknown Class ', xml_document, class_name)
         #print(f'{xml_document.split("/")[-1]} {class_name}')
 
 
 def generate_txt_from_xml():
-    class_distribution = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
+    class_distribution = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,0,0]
 
     filepaths = glob(join(annotations , '*.xml'))
 

our_scripts/run_yolov3_process.py

@@ -1,4 +1,3 @@
-import argparse
 import datetime
 import glob
 import io
@@ -12,22 +11,25 @@ from config import Configuration
 
 def call_training_script(config):
     cmd = '/home/tomekb/miniconda3/envs/conda3.7/bin/python -u /home/tomekb/yolov3/train.py '
-    cmd += config.train.get_args_string()
+    cmd += f"--experiment-dir {config.experiments.dir}"
+    cmd += config.train.get_args_string()  # getting rest of train arguments
+
     print("_______ CALLING TRAINING SCRIPT _______")
     print(cmd)
 
-    os.chdir('..')
+    os.chdir('..')  # change to project root directory
     process = subprocess.Popen(cmd, stdout=subprocess.PIPE, shell=True)
     for line in io.TextIOWrapper(process.stdout, encoding="utf-8"):  # print output of training process to console
         print(line)
     return cmd
 
+
 def move_training_results_to_experiments_dir(config):
-    training_results_dir_path = os.path.join(config.experiments.dir, datetime.datetime.now().strftime('%Y-%m-%d_%H-%M-%S'))
+    training_results_dir_path = os.path.join(config.experiments.dir, datetime.datetime.now().strftime('%Y-%m-%d_%H-%M-%S')) #creating directory accordint to pattern eg: 2020-06-30_17-52-19
+
     print("_______ CALLING MOVING RESULTS _______")
     print(f"MOVING RESUTLS TO {training_results_dir_path}")
 
-
     os.mkdir(training_results_dir_path)
 
     weights_path = os.path.join(training_results_dir_path, 'best.pt')
@@ -36,14 +38,14 @@ def move_training_results_to_experiments_dir(config):
     names_path = open(config.train.data).readlines()[3].split('=')[-1].rstrip()  # read names path from file
     names_file_name = ntpath.basename(names_path)
     experiment_names_path = os.path.join(training_results_dir_path, names_file_name)
-    shutil.copy(names_path, experiment_names_path)                                      # copy names to created experiment dir with training results
+    shutil.copy(names_path, experiment_names_path)  # copy names file from *.data file to created experiment dir with training results
 
     tensorboard_dir = './runs'
     last_modified_tensorboard_dir = max(glob.glob(os.path.join(tensorboard_dir, '*/')), key=os.path.getmtime)
     shutil.move(last_modified_tensorboard_dir, os.path.join(training_results_dir_path))  # saving related tensorboard dir
 
-    shutil.copy2(config.config_path, training_results_dir_path)         #copying configuration yaml
 
+    shutil.copy2(config.config_path, training_results_dir_path)  # copying configuration yaml
 
     # for test purposes only
     # shutil.copy2('/home/tomekb/yolov3/experiments/1/best.pt', training_results_dir_path)
@@ -69,7 +71,6 @@ def call_detection_script(config, weights_path, names_path, dir):
 
     cmd = " ".join(cmd.split())
 
-
     print("_______ CALLING DETECTION SCRIPT _______")
     print(cmd)
 

our_scripts/run_yolov3_process_bayes.py

@@ -1,31 +1,56 @@
 import datetime
 import glob
-import io
 import ntpath
 import os
 import shutil
-import subprocess
+import traceback
 
-from .config import Configuration
+import GPyOpt
+import numpy as np
+
+from config_bayes import Configuration
+from utils import call_subprocess, get_values_from_conff_matrix
+
+config = Configuration()
+
+date_string = datetime.datetime.now().strftime('%Y-%m-%d_%H-%M-%S')
+bayes_params_file = open(os.path.join(config.experiments.dir, f"{date_string}_bayes_params.txt"), 'a+')
 
 
-def call_training_script(config):
+def call_training_script(gaussian_hyps):
-    cmd = '/home/tomekb/miniconda3/envs/conda3.7/bin/python -u /home/tomekb/yolov3/train.py '
+    cmd = 'python-u /home/tomekb/yolov3/train.py'
-    cmd += f"--experiment-dir {config.experiments.dir}"
+    cmd += ' --epochs ' + gaussian_hyps['epochs'].__str__()
-    cmd += config.train.get_args_string()  # getting rest of train arguments
+    cmd += ' --batch-size ' + gaussian_hyps['batch-size'].__str__()
+    cmd += ' --cfg ' + config.train.cfg.__str__()
+    cmd += ' --data ' + config.train.data.__str__()
+    cmd += ' --multi-scale ' if gaussian_hyps['multi-scale'] else ""
+    cmd += ' --img-size ' + gaussian_hyps['img-size']
+    cmd += ' --rect ' if gaussian_hyps['rect'] else ""
+    cmd += ' --weights ' + config.train.weights.__str__()
+    cmd += ' --device ' + config.train.device.__str__()
+    cmd += ' --adam ' if gaussian_hyps['adam'] else ""
+    cmd += ' --freeze-layers ' if getattr(config.train, "freeze-layers") else ""
+    # cmd += ' --snapshot-every ' if getattr(config.train, "snapshot-every") else ""
+    cmd += ' --experiment-dir ' + config.experiments.dir.__str__()
+
+    train_hyps = dict(
+        (key, gaussian_hyps[key]) for idx, (key, _) in enumerate(gaussian_hyps.items()) if idx in range(6, 24))
+    cmd += f' --hyp \"{train_hyps}\"'
 
     print("_______ CALLING TRAINING SCRIPT _______")
     print(cmd)
 
-    os.chdir('..')  # change to project root directory
+    dir_path = os.path.dirname(os.path.realpath(__file__))
-    process = subprocess.Popen(cmd, stdout=subprocess.PIPE, shell=True)
+    os.chdir(os.path.join(dir_path, '..'))  # change to project root directory
-    for line in io.TextIOWrapper(process.stdout, encoding="utf-8"):  # print output of training process to console
+
-        print(line)
+    call_subprocess(cmd)
+
     return cmd
 
 
-def move_training_results_to_experiments_dir(config):
+def move_training_results_to_experiments_dir():
-    training_results_dir_path = os.path.join(config.experiments.dir, datetime.datetime.now().strftime('%Y-%m-%d_%H-%M-%S')) #creating directory accordint to pattern eg: 2020-06-30_17-52-19
+    training_results_dir_path = os.path.join(config.experiments.dir, datetime.datetime.now().strftime(
+        '%Y-%m-%d_%H-%M-%S'))  # creating directory accordint to pattern eg: 2020-06-30_17-52-19
 
     print("_______ CALLING MOVING RESULTS _______")
     print(f"MOVING RESUTLS TO {training_results_dir_path}")
@@ -38,22 +63,26 @@ def move_training_results_to_experiments_dir(config):
     names_path = open(config.train.data).readlines()[3].split('=')[-1].rstrip()  # read names path from file
     names_file_name = ntpath.basename(names_path)
     experiment_names_path = os.path.join(training_results_dir_path, names_file_name)
-    shutil.copy(names_path, experiment_names_path)  # copy names file from *.data file to created experiment dir with training results
+    shutil.copy(names_path,
-
+                experiment_names_path)  # copy names file from *.data file to created experiment dir with training results
-    tensorboard_dir = './runs'
-    last_modified_tensorboard_dir = max(glob.glob(os.path.join(tensorboard_dir, '*/')), key=os.path.getmtime)
-    shutil.move(last_modified_tensorboard_dir, os.path.join(training_results_dir_path))  # saving related tensorboard dir
 
+    tensorboard_dir = '/home/tomekb/yolov3/runs'
+    tensorboard_events_files = glob.glob(os.path.join(tensorboard_dir, '*'))
+    last_modified_events_file = max(tensorboard_events_files, key=os.path.getmtime)
+    shutil.move(last_modified_events_file,
+                os.path.join(training_results_dir_path))  # saving related tensorboard dir
 
     shutil.copy2(config.config_path, training_results_dir_path)  # copying configuration yaml
 
     # for test purposes only
-    # shutil.copy2('/home/tomekb/yolov3/experiments/1/best.pt', training_results_dir_path)
+    # TODO CHANGE ME AFTER TESTS
+    shutil.copy2('/home/tomekb/yolov3/experiments/yolov3-spp-100-epochs-freeze-layers/best.pt',
+                 training_results_dir_path)
 
     return weights_path, experiment_names_path, training_results_dir_path
 
 
-def call_detection_script(config, weights_path, names_path, dir):
+def call_detection_script(gaussian_hyps, weights_path, names_path, dir):
     detect_output_dir = os.path.join(dir, 'output')
     cmd = f"""/home/tomekb/miniconda3/envs/conda3.7/bin/python -u /home/tomekb/yolov3/detect.py
           --cfg {config.train.cfg}
@@ -61,41 +90,101 @@ def call_detection_script(config, weights_path, names_path, dir):
           --output {detect_output_dir}
           --names {names_path}
           --weights {weights_path}
-          --test-img-size {getattr(config.detect, 'test-img-size')}
+          --test-img-size {gaussian_hyps['test-img-size']}
-          --conf-thres {getattr(config.detect, 'conf-thres')}
+          --conf-thres {gaussian_hyps['conf-thres']}
-          --iou-thres {getattr(config.detect, 'iou-thres')}
+          --iou-thres {gaussian_hyps['iou-thres']}
           --save-txt"""
-    cmd += " --agnostic-nms" if getattr(config.detect, 'agnostic-nms') else ""
-    cmd += " --agument" if getattr(config.detect, 'augment') else ""
     cmd += f" --device {config.train.device}" if config.train.device else ""
-
     cmd = " ".join(cmd.split())
 
     print("_______ CALLING DETECTION SCRIPT _______")
     print(cmd)
 
-    process = subprocess.Popen(cmd, stdout=subprocess.PIPE, shell=True)
+    call_subprocess(cmd)
-    for line in io.TextIOWrapper(process.stdout, encoding="utf-8"):  # print output of  process to console
-        print(line)
-
     return detect_output_dir
 
 
-def call_generate_confussion_matrix(detect_output_dir, config, names_path, train_results_dir):
+def call_generate_confussion_matrix(detect_output_dir, names_path, train_results_dir):
     labels_dir = getattr(config.confussion_matrix, 'labels-dir')
-
+    conff_matrix_path = os.path.join(train_results_dir, 'confussion-matrix.tsv')
-    cmd = f"node ./our_scripts/generate-confusion-matrix.js {detect_output_dir} {labels_dir} {names_path} > {train_results_dir}/confussion-matrix.tsv"
+    cmd = f"node /home/tomekb/yolov3/our_scripts/generate-confusion-matrix.js {detect_output_dir} {labels_dir} {names_path} > {conff_matrix_path}"
-    process = subprocess.Popen(cmd, stdout=subprocess.PIPE, shell=True)
     print("_______ CALLING CONFUSSION MATRIX SCRIPT _______")
     print(cmd)
-    for line in io.TextIOWrapper(process.stdout, encoding="utf-8"):  # print output of  process to console
+    call_subprocess(cmd)
-        print(line)
+    return conff_matrix_path
 
 
+def yolov3(x):
+    bayes_hyps = {
+        'epochs': int(x[:, 0]),
+        'batch-size': int(x[:, 1]),
+        'multi-scale': bool(x[:, 2]),
+        'img-size': f"{int(x[:, 3])} {int(x[:, 4])}",
+        'rect': bool(x[:, 5]),
+        'adam': bool(x[:, 6]),
+        'giou': float(x[:, 7]),  # train hyps start index
+        'cls': float(x[:, 8]),
+        'cls_pw': float(x[:, 9]),
+        'obj': float(x[:, 10]),
+        'obj_pw': float(x[:, 11]),
+        'iou_t': float(x[:, 12]),
+        'lr0': float(x[:, 13]),
+        'lrf': float(x[:, 14]),
+        'momentum': float(x[:, 15]),
+        'weight_decay': float(x[:, 16]),
+        'fl_gamma': float(x[:, 17]),
+        'hsv_h': float(x[:, 18]),
+        'hsv_s': float(x[:, 19]),
+        'hsv_v': float(x[:, 20]),
+        'degrees': float(x[:, 21]),
+        'translate': float(x[:, 22]),
+        'scale': float(x[:, 23]),
+        'shear': float(x[:, 24]),  # train hyps end index
+        'test-img-size': int(x[:, 25]),
+        'conf-thres': float(x[:, 26]),
+        'iou-thres': float(x[:, 27])
+    }
+
+    line = ""
+    try:
+        call_training_script(bayes_hyps)
+        weights_path, names_path, train_results_dir = move_training_results_to_experiments_dir()
+        detect_output_dir = call_detection_script(bayes_hyps, weights_path, names_path, train_results_dir)
+        conf_matrix_path = call_generate_confussion_matrix(detect_output_dir, names_path, train_results_dir)
+
+        y_dict = get_values_from_conff_matrix(conf_matrix_path)
+
+        # tutaj wzór na wyliczanie funkcji 
+        y_val = 1 - ((y_dict['match'] * 10 - y_dict['false positives'] * 3) / y_dict['mistakes'])
+
+        # zapisywanie do pliku zadeklarowanego globalnie
+        line = "\t".join([bayes_hyps.__str__(), str(y_val)])
+        bayes_params_file.writelines([line, '\n'])
+        return y_val
+    except:
+        tb = traceback.format_exc()
+        print("An error occured during running training-detect-confussion process \n", tb)
+        print("Returning 1 from current bayessian iteration")
+        line = "\t".join([bayes_hyps.__str__(), str(1)])
+        return 1
+    finally:
+        bayes_params_file.writelines([line, '\n'])
+
+
+
+# uruchamiać z
 if __name__ == '__main__':
-    config = Configuration()
 
-    train_cmd = call_training_script(config)
+    bounds = config.get_bayes_bounds()
-    weights_path, names_path, train_results_dir = move_training_results_to_experiments_dir(config)
+
-    detect_output_dir = call_detection_script(config, weights_path, names_path, train_results_dir)
+    # for b in bounds:
-    call_generate_confussion_matrix(detect_output_dir, config, names_path, train_results_dir)
+    #   print(b)
+
+    # tutaj będzie wczytywanie z poprzednich eksperymentów plik bayes_params
+    X = None
+    Y = None
+
+    bayes_optimizer = GPyOpt.methods.BayesianOptimization(f=yolov3, domain=bounds, X=X, Y=Y, verbosity=True,
+                                                          initial_design_numdata=2)
+    bayes_optimizer.run_optimization(config.bayes.iterations, verbosity=True)
+    bayes_params_file.close()

our_scripts/utils.py

@@ -0,0 +1,55 @@
+import io
+import subprocess
+
+
+def call_subprocess(cmd):
+    process = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, shell=True)
+    error = False
+    for process_line_output in io.TextIOWrapper(process.stdout,
+                                                encoding="utf-8"):  # print output of training process to console
+        if 'Traceback' in process_line_output:
+            error = True
+        print(process_line_output)
+    if error:
+        raise RuntimeError("An error occured during calling subprocess")
+
+
+def get_values_from_conff_matrix(path):
+    lines = open(path, 'r').readlines()[:7]
+    d = {}
+    for l in lines:
+        key, value, *_ = l.split("\t")
+        d.update({key.replace(":", ""): int(value)})
+    return d
+
+
+def get_bayes_params_as_dict(x):
+    return {
+        'epochs': int(x[:, 0]),
+        'batch-size': int(x[:, 1]),
+        'multi-scale': bool(x[:, 2]),
+        'img-size': f"{int(x[:, 3])} {int(x[:, 4])}",
+        'rect': bool(x[:, 5]),
+        'adam': bool(x[:, 6]),
+        'giou': float(x[:, 7]),  # train hyps start index
+        'cls': float(x[:, 8]),
+        'cls_pw': float(x[:, 9]),
+        'obj': float(x[:, 10]),
+        'obj_pw': float(x[:, 11]),
+        'iou_t': float(x[:, 12]),
+        'lr0': float(x[:, 13]),
+        'lrf': float(x[:, 14]),
+        'momentum': float(x[:, 15]),
+        'weight_decay': float(x[:, 16]),
+        'fl_gamma': float(x[:, 17]),
+        'hsv_h': float(x[:, 18]),
+        'hsv_s': float(x[:, 19]),
+        'hsv_v': float(x[:, 20]),
+        'degrees': float(x[:, 21]),
+        'translate': float(x[:, 22]),
+        'scale': float(x[:, 23]),
+        'shear': float(x[:, 24]),  # train hyps end index
+        'test-img-size': int(x[:, 25]),
+        'conf-thres': float(x[:, 26]),
+        'iou-thres': float(x[:, 27])
+    }

train.py

@@ -10,6 +10,7 @@ from models import *
 from utils.datasets import *
 from utils.utils import *
 from our_scripts.config import  Configuration
+import ast
 
 
 mixed_precision = True
@@ -117,7 +118,6 @@ def train(hyp):
     best_fitness = 0.0
     attempt_download(weights)
     if weights.endswith('.pt'):  # pytorch format
-        print("LOADIN MODEL")
         # possible weights are '*.pt', 'yolov3-spp.pt', 'yolov3-tiny.pt' etc.
         ckpt = torch.load(weights, map_location=device)
 
@@ -364,8 +364,9 @@ def train(hyp):
                          'model': ema.ema.module.state_dict() if hasattr(model, 'module') else ema.ema.state_dict(),
                          'optimizer': None if final_epoch else optimizer.state_dict()}
 
-            if epoch % opt.save_every_nth_epoch == 0:
+            if opt.snapshot_every and epoch % opt.snapshot_every == 0 :
-                torch.save(chkpt, f'yolo_{epoch}.pt')
+                saving_path = os.path.join(opt.experiment_dir, f'weights_{epoch}.pt')
+                torch.save(ckpt, saving_path)
             # Save last, best and delete
             torch.save(ckpt, last)
             if (best_fitness == fi) and not final_epoch:
@@ -416,8 +417,14 @@ if __name__ == '__main__':
     parser.add_argument('--adam', action='store_true', help='use adam optimizer')
     parser.add_argument('--single-cls', action='store_true', help='train as single-class dataset')
     parser.add_argument('--freeze-layers', action='store_true', help='Freeze non-output layers')
-    parser.add_argument('--save-every-nth-epoch', type=int, help='Saving every n-th epoth')
+
+    # parametry dodane na cele projektu wykrywania aut
+    parser.add_argument('--snapshot-every', type=int, help='Saving every n-th state of model weights')
+    parser.add_argument('--experiment-dir', type=str, help='Directory for experiments')
+    parser.add_argument('--hyp', type=str, help='String that represents dictionary with hyperparameters ')
+
     opt = parser.parse_args()
+    #print(opt)
     #opt.weights = last if opt.resume and not opt.weights else opt.weights
     #check_git_status()
     opt.cfg = check_file(opt.cfg)  # check file
@@ -431,9 +438,10 @@ if __name__ == '__main__':
     # scale hyp['obj'] by img_size (evolved at 320)
     # hyp['obj'] *= opt.img_size[0] / 320.
 
-
+    #overriding global hyp variable with our bayessian hyps
-    hyp = Configuration().train.other_hyps.__dict__
+    hyp = ast.literal_eval(opt.hyp)
-
+    #print('### TRAIN HYPERPARAMETERS ###')
+    #print(hyp)
 
     tb_writer = None
     if not opt.evolve:  # Train normally

utils/datasets.py

@@ -553,6 +553,7 @@ def augment_hsv(img, hgain=0.5, sgain=0.5, vgain=0.5):
     img_hsv = cv2.merge((cv2.LUT(hue, lut_hue), cv2.LUT(sat, lut_sat), cv2.LUT(val, lut_val))).astype(dtype)
     cv2.cvtColor(img_hsv, cv2.COLOR_HSV2BGR, dst=img)  # no return needed
     
+
     # Histogram equalization
     # if random.random() < 0.2:
     #     for i in range(3):