v2

.gitignore

@@ -155,3 +155,4 @@ node_modules/
 /test/yolov7-tracker/runs
 /test/yolov7-tracker/wandb
 VoTT-v2-none.zip
+VoTT-v2-done.zip

yolov7-setup/v2/compute_yolov3_anchors_v2.py

@@ -0,0 +1,96 @@
+import json
+import os, sys
+import math
+import matplotlib.pyplot as plt
+import numpy as np
+import seaborn as sns; sns.set()  # for plot styling
+from PIL import Image
+
+imgsize = int(input("Image size for train and run? "))
+print("=> imgsize =",imgsize)
+# imgsize = 256
+
+BBDlabeldict = {"runner":0}
+
+datasetpath = '/Users/noham/Documents/GitHub/Stage-2024/yolov7-setup/v2/datasetv2/'
+images_path = datasetpath+"images"
+labels_path = datasetpath+"labels"
+
+filenames = sorted(os.listdir(images_path))
+print("Nb d'images =",len(filenames))
+w,h = [] , []
+for i,file in enumerate(filenames):
+    #print(file)
+    try:
+        # if True:
+        im = Image.open(images_path+"/"+file)
+        img_width = im.width
+        img_height = im.height
+        #print(img_width,img_height)
+        fo = open(images_path.replace("images","labels")+"/"+file.replace("jpg","txt"), "r+")
+        lines = fo.readlines()
+        for line in lines:
+            a = np.array(line.split(" ")).astype(float)
+            w.append(a[3]*img_width)
+            h.append(a[4]*img_height)
+            if (a[3]*img_height<0.001) or (a[4]*img_height<0.001):
+                print("!! ATTENTION : boite trop petite dans le fichier ",file, "=> il faut vérifier et supprimer la ligne dans le fichier label au format txt !!")
+    except:
+        pass
+w=np.asarray(w)#+0.001*(np.random.random((len(w),)) - 0.5)
+h=np.asarray(h)#+0.001*(np.random.random((len(w),)) - 0.5)
+print("h => min =",h.min()," max =",h.max())
+print("w => min =",w.min()," max =",h.max())
+print("Nb objets entourés =",len(h))
+
+x=[w,h]
+x=np.asarray(x)
+x=x.transpose()
+##########################################   K- Means
+##########################################
+
+from sklearn.cluster import KMeans
+kmeans3 = KMeans(n_clusters=9)
+kmeans3.fit(x)
+y_kmeans3 = kmeans3.predict(x)
+
+##########################################
+centers3 = kmeans3.cluster_centers_
+
+yolo_anchor_average=[]
+for ind in range (9):
+    yolo_anchor_average.append(np.mean(x[y_kmeans3==ind],axis=0))
+
+yolo_anchor_average=np.array(yolo_anchor_average)
+
+plt.scatter(x[:, 0], x[:, 1], c=y_kmeans3, s=2, cmap='viridis')
+plt.scatter(yolo_anchor_average[:, 0], yolo_anchor_average[:, 1], c='red', s=50)
+yoloV3anchors = yolo_anchor_average
+yoloV3anchors[:, 0] =yolo_anchor_average[:, 0] /1920 *imgsize
+yoloV3anchors[:, 1] =yolo_anchor_average[:, 1] /1056 *imgsize
+yoloV3anchors = np.rint(yoloV3anchors)
+fig, ax = plt.subplots()
+for ind in range(9):
+    rectangle= plt.Rectangle((0.5*imgsize-yoloV3anchors[ind,0]/2,0.5*imgsize-yoloV3anchors[ind,1]/2), yoloV3anchors[ind,0],yoloV3anchors[ind,1] , fc='b',edgecolor='b',fill = None)
+    ax.add_patch(rectangle)
+ax.set_aspect(1.0)
+plt.axis([0,imgsize,0,imgsize])
+plt.savefig(datasetpath+"boites.png",dpi=150)
+plt.show()
+
+yoloV3anchors.sort(axis=0)
+print("Your custom anchor boxes are {}".format(yoloV3anchors))
+
+F = open(datasetpath+"YOLOV_BDD_Anchors_"+str(imgsize)+".txt", "w")
+F.write("{}".format(yoloV3anchors))
+F.close()
+
+print("Anchors box for yaml file :")
+s_anchors = yoloV3anchors[yoloV3anchors[:, 0].argsort()]
+anchor_lists = [s_anchors[i:i+3].tolist() for i in range(0, len(s_anchors), 3)]
+# anchor_lists = [[[14.0, 37.0], [15.0, 48.0], [19.0, 53.0]], [[21.0, 70.0], [24.0, 88.0], [31.0, 112.0]], [[34.0, 136.0], [35.0, 155.0], [76.0, 161.0]]]
+p = [']  # P3/8', ']  # P4/16', ']  # P5/32']
+out = "anchors:"
+for l in anchor_lists:
+    out += '\n' + "  - [" + ", ".join([f"{int(i[0])},{int(i[1])}" for i in l]) + p[anchor_lists.index(l)]
+print(out)

yolov7-setup/v2/datasetv2/YOLOV_BDD_Anchors_720.txt

@@ -0,0 +1,9 @@
+[[ 23.  58.]
+ [ 27. 104.]
+ [ 38. 106.]
+ [ 46. 146.]
+ [ 47. 156.]
+ [ 68. 192.]
+ [ 68. 232.]
+ [134. 273.]
+ [204. 498.]]

yolov7-setup/v2/datasetv2/dataset.yaml

@@ -0,0 +1,4 @@
+train: data/datasetv2/liste_images.txt
+val: data/datasetv2/liste_images.txt
+nc: 1
+names: ['runner']