compute_seg_result.py

+from ultralytics import YOLO
+import cv2
+import matplotlib.pyplot as plt
+import argparse
+import os
+
+# Create the parser
+parser = argparse.ArgumentParser()
+# Add an argument
+parser.add_argument('--path_evaluation', type=str, required=True)
+parser.add_argument('--path_result', type=str, required=True)
+parser.add_argument('--class_object', type=str, required=True)
+parser.add_argument('--path_model_yolo', type=str, required=True)
+# Parse the argument
+args = parser.parse_args()
+
+
+# load the model.
+model = YOLO(f"{args.path_model_yolo}")
+#
+
+path_evaluation_data = args.path_evaluation + "/" + args.class_object + "/RGB_resized"
+
+for files in os.listdir(path_evaluation_data): 
+    #print("files : ", files)
+    #print(f"{path_evaluation_data}/{files}")
+
+    try:
+        results = model.predict(source=f"{path_evaluation_data}/{files}", conf=0.5, 
+save=True)
+    #print(results)
+        results1 = results[0].to('cpu')
+        results11 = results1.numpy()
+    # boxes = results11.boxes  # Boxes object for bbox outputs
+    # probs = results11.probs  # Class probabilities for classification outputs
+#
+        masks = results11.masks  # Masks object for segmentation masks outputs
+        #print(masks)
+        mask_res = masks.data[0]
+        plt.imsave(f"{args.path_result}/{files}", mask_res, cmap='gray') 
+        print("images saved : ", files )
+    except : 
+        print("no prediction")

datasets/linemod/dataset.py

@@ -24,12 +24,11 @@ import matplotlib.pyplot as plt
 
 class PoseDataset(data.Dataset):
     def __init__(self, mode, num, add_noise, root, noise_trans, refine):
-        # ["banana1", "kiwi1", "pear2", "strawberry1", "apricot", "orange2", "peach1", "lemon2", "apple2" ]
         #self.objlist = [1] #TODO
 
-        # apple, ....
-        #self.objlist = [1, 2, 3, 4, 5, 6, 7, 8]
-        self.objlist = [1]
+        # apple2 0 , apricot 1 , banana1 2 , kiwi1 3 , lemon2 4 , orange2 5 , peach1 6 , pear2 7
+        self.objlist = [1, 2, 3, 4, 5, 6, 7, 8]
+        #self.objlist = [1, 3, 6, 7, 8]
         self.mode = mode
 
         self.list_rgb = []
@@ -94,8 +93,8 @@ class PoseDataset(data.Dataset):
         self.border_list = [-1, 40, 80, 120, 160, 200, 240, 280, 320, 360, 400, 440, 480, 520, 560, 600, 640, 680]
         self.num_pt_mesh_large = 500
         self.num_pt_mesh_small = 500
-        self.symmetry_obj_idx = [1] # TODO
-        #self.symmetry_obj_idx = [0 ,1, 3, 4, 5, 6]
+        self.symmetry_obj_idx = [0, 1, 3, 4, 5, 6] # TODO
+        #self.symmetry_obj_idx = [0, 2, 3]
 
 
     def __getitem__(self, index):
@@ -130,8 +129,8 @@ class PoseDataset(data.Dataset):
 
         mask_depth = ma.getmaskarray(ma.masked_not_equal(depth, 0))
         if self.mode == 'eval':
-            #mask_label = ma.getmaskarray(ma.masked_equal(label, np.array(255)))
-            mask_label = ma.getmaskarray(ma.masked_equal(label, np.array([255, 255, 255])))[:, :, 0]
+            mask_label = ma.getmaskarray(ma.masked_equal(label, np.array(255)))
+            #mask_label = ma.getmaskarray(ma.masked_equal(label, np.array([255, 255, 255])))[:, :, 0]
         else: 
             #mask_label = ma.getmaskarray(ma.masked_equal(label, np.array([255, 255, 255])))[:, :, 0]
             mask_label = ma.getmaskarray(ma.masked_equal(label, np.array(255)))
@@ -193,6 +192,10 @@ class PoseDataset(data.Dataset):
 
         model_points = self.pt[obj] / 1000.0
         dellist = [j for j in range(0, len(model_points))]
+        #print("del :", len(dellist))
+        #print("len : ", len(model_points)-self.num_pt_mesh_small)
+        #print("model point : ", len(model_points))
+        #print("num_pt_mesh : ", self.num_pt_mesh_small)
         dellist = random.sample(dellist, len(model_points) - self.num_pt_mesh_small)
         model_points = np.delete(model_points, dellist, axis=0)
 

datasets/linemod/dataset_config/models_info.yml

-1: {diameter: 52.94260215554199, min_x: -26.221, min_y: -23.598, min_z: -25.405, size_x: 52.2, size_y: 47.816, size_z: 51.055},
-2: {diameter: 55.64182762994041, min_x: -21.122, min_y: -28.445, min_z: -22.181, size_x: 42.491, size_y: 55.287, size_z: 43.699},
-3: {diameter: 152.6314929798569, min_x: -14.428, min_y: -75.92699, min_z: -23.056, size_x: 29.562, size_y: 151.43699, size_z: 60.727000000000004},
-4: {diameter: 72.65268354988685, min_x: -24.562, min_y: -36.235, min_z: -24.574, size_x: 49.108000000000004, size_y: 72.094, size_z: 49.120000000000005},
-5: {diameter: 72.58068124508065, min_x: -23.86, min_y: -23.435, min_z: -35.375, size_x: 46.884, size_y: 46.866, size_z: 72.48400000000001},
-6: {diameter: 77.41278909198401, min_x: -36.512,min_y: -36.759, min_z: -37.344, size_x: 73.693, size_y: 76.06, size_z: 74.78200000000001},
-7: {diameter: 76.63639853229013, min_x: -37.55, min_y: -35.435, min_z: -36.938, size_x: 74.026, size_y: 71.15899999999999, size_z: 76.622},
+1: {diameter: 52.94260215554199, min_x: -26.221, min_y: -23.598, min_z: -25.405, size_x: 52.2, size_y: 47.816, size_z: 51.055}
+2: {diameter: 55.64182762994041, min_x: -21.122, min_y: -28.445, min_z: -22.181, size_x: 42.491, size_y: 55.287, size_z: 43.699}
+3: {diameter: 152.6314929798569, min_x: -14.428, min_y: -75.92699, min_z: -23.056, size_x: 29.562, size_y: 151.43699, size_z: 60.727000000000004}
+4: {diameter: 72.65268354988685, min_x: -24.562, min_y: -36.235, min_z: -24.574, size_x: 49.108000000000004, size_y: 72.094, size_z: 49.120000000000005}
+5: {diameter: 72.58068124508065, min_x: -23.86, min_y: -23.435, min_z: -35.375, size_x: 46.884, size_y: 46.866, size_z: 72.48400000000001}
+6: {diameter: 77.41278909198401, min_x: -36.512,min_y: -36.759, min_z: -37.344, size_x: 73.693, size_y: 76.06, size_z: 74.78200000000001}
+7: {diameter: 76.63639853229013, min_x: -37.55, min_y: -35.435, min_z: -36.938, size_x: 74.026, size_y: 71.15899999999999, size_z: 76.622}
 8: {diameter: 129.85637252518683, min_x: -32.724, min_y: -52.947, min_z: -33.292, size_x: 66.189, size_y: 129.07999999999998, size_z: 67.57300000000001}

experiments/scripts/eval_linemod.sh

@@ -7,5 +7,4 @@ export PYTHONUNBUFFERED="True"
 export CUDA_VISIBLE_DEVICES=0
 
 python3 ./tools/eval_linemod.py --dataset_root ./datasets/linemod/Linemod_preprocessed\
-  --model trained_checkpoints/linemod/pose_model_9_0.01310166542980859.pth\
-  --refine_model trained_checkpoints/linemod/pose_refine_model_493_0.006761023565178073.pth
\ No newline at end of file
+  --model trained_models/linemod8/pose_model_4_0.012983739659874712.pth --refine_model trained_models/linemod8/pose_refine_model_9_0.01186443073513208.pth

slurm/DF2.slurm

+#!/bin/bash
+#SBATCH --job-name=DF # nom du job
+#SBATCH -A uli@gpu
+#SBATCH --ntasks=1 # Nombre de processus MPI
+#SBATCH --cpus-per-task=10 # nombre de threads OpenMP
+#SBATCH --hint=nomultithread # pas d'hyperthreading
+#SBATCH --qos=qos_gpu-t4
+#SBATCH --gres=gpu:1
+#SBATCH --time=100:00:00 # Temps d’exécution max
+#SBATCH --output=Output/Pvnet%j.out # fichier de sortie
+#SBATCH --error=Output/Pvnet%j.out # fichier d'erreur
+
+# on se place dans le répertoire de soumission
+cd ${SLURM_SUBMIT_DIR}
+
+module purge
+# chargement des modules
+
+module load python/3.6.15
+
+# echo des commandes lancées
+set -x
+
+
+cat_target=$1
+echo $cat_target
+
+
+conda activate NewDF
+bash ./experiments/scripts/train_linemod.sh
+

tools/eval_linemod.py

@@ -132,8 +132,10 @@ parser.add_argument('--model', type=str, default='', help='resume PoseNet model'
 parser.add_argument('--refine_model', type=str, default='', help='resume PoseRefineNet model')
 opt = parser.parse_args()
 
+# num_objects = 5
 num_objects = 8
-objlist = [1, 2, 3, 4, 5, 6, 7, 8]
+# objlist = [1, 3, 6, 7, 8]
+objlist = [1, 2, 3, 4 ,5 ,6, 7, 8]
 num_points = 500
 iteration = 4
 bs = 1
@@ -143,7 +145,7 @@ cam_fx = 543.2527222420504  # TODO
 cam_fy = 724.3369629894005  # TODO
 
 # ["banana1", "kiwi1", "pear2", "strawberry1", "apricot", "orange2", "peach1", "lemon2", "apple2" ]
-map_id_obj = {
+"""map_id_obj = {
     1: 'banana1',
     2: 'kiwi1',
     3: 'pear2',
@@ -152,7 +154,14 @@ map_id_obj = {
     6: 'peach1',
     7: 'lemon2',
     8: 'apple2',
-}
+}"""
+#map_id_obj = {1: 'apple2', 3: 'banana1'}
+map_id_obj = {1: 'apple2', 2: 'apricot', 3: 'banana1', 4: 'kiwi1', 5:'lemon2',  6: 'orange2', 7: 'peach1', 8: 'pear2'}
+        # apple2, apricot, banana1, kiwi1, lemon2, orange2, peach1, pear2
+        #self.objlist = [1, 2, 3, 4, 5, 6, 7, 8]
+        #self.objlist = [1, 3, 6, 7, 8]
+
+
 K = np.array([[cam_fx, 0, cam_cx], [0, cam_fy, cam_cy], [0, 0, 1]])
 dataset_config_dir = 'datasets/linemod/dataset_config'
 output_result_dir = 'experiments/eval_result/linemod'
@@ -176,8 +185,9 @@ criterion = Loss(num_points_mesh, sym_list)
 criterion_refine = Loss_refine(num_points_mesh, sym_list)
 
 diameter = []
+print('{0}/models_info.yml'.format(dataset_config_dir))
 meta_file = open('{0}/models_info.yml'.format(dataset_config_dir), 'r')
-meta = yaml.load(meta_file)
+meta = yaml.load(meta_file, yaml.Loader)
 for obj in objlist:
     diameter.append(meta[obj]['diameter'] / 1000.0 * 0.1)
 print(diameter)

tools/train.py

@@ -30,7 +30,7 @@ from lib.utils import setup_logger
 parser = argparse.ArgumentParser()
 parser.add_argument('--dataset', type=str, default = 'ycb', help='ycb or linemod')
 parser.add_argument('--dataset_root', type=str, default = '', help='dataset root dir (''YCB_Video_Dataset'' or ''Linemod_preprocessed'')')
-parser.add_argument('--batch_size', type=int, default = 8, help='batch size')
+parser.add_argument('--batch_size', type=int, default = 32, help='batch size')
 parser.add_argument('--workers', type=int, default = 10, help='number of data loading workers')
 parser.add_argument('--lr', default=0.0001, help='learning rate')
 parser.add_argument('--lr_rate', default=0.3, help='learning rate decay rate')
@@ -40,7 +40,7 @@ parser.add_argument('--decay_margin', default=0.016, help='margin to decay lr &
 parser.add_argument('--refine_margin', default=0.013, help='margin to start the training of iterative refinement')
 parser.add_argument('--noise_trans', default=0.03, help='range of the random noise of translation added to the training data')
 parser.add_argument('--iteration', type=int, default = 2, help='number of refinement iterations')
-parser.add_argument('--nepoch', type=int, default=500, help='max number of epochs to train')
+parser.add_argument('--nepoch', type=int, default=60, help='max number of epochs to train')
 parser.add_argument('--resume_posenet', type=str, default = '',  help='resume PoseNet model')
 parser.add_argument('--resume_refinenet', type=str, default = '',  help='resume PoseRefineNet model')
 parser.add_argument('--start_epoch', type=int, default = 1, help='which epoch to start')
@@ -59,12 +59,12 @@ def main():
         opt.log_dir = 'experiments/logs/ycb' #folder to save logs
         opt.repeat_epoch = 1 #number of repeat times for one epoch training
     elif opt.dataset == 'linemod':
-        #opt.num_objects = 8 #TODO
-        opt.num_objects = 1 #TODO
+        opt.num_objects = 8 #TODO
+        #opt.num_objects = 5 #TODO
         opt.num_points = 500
-        opt.outf = 'trained_models/linemod'
-        opt.log_dir = 'experiments/logs/linemod'
-        opt.repeat_epoch = 20
+        opt.outf = 'trained_models/linemod8'
+        opt.log_dir = 'experiments/logs/linemod8'
+        opt.repeat_epoch = 5
     else:
         print('Unknown dataset')
         return