def eval_train(checkpoint_dir, checkpoint_list, image_set):
'''Evaluate training on list of stored checkpoints'''
results = dict()
for cp in checkpoint_list:
cp_path = checkpoint_dir + 'model.ckpt-' + str(cp)
print('Evaluating: ' + cp_path)
net = nn.init(cp_path)
ap, recall = eval_set(net, image_set)
results[cp] = (ap, recall)
return results
return cp, np.mean(all_ap, axis=0), np.mean(all_rec, axis=0) #CHECKPOINT = '/home/tommaso/scenicEx/data/scenic/checkpoints/train/model.ckpt-5500' #CHECKPOINT = '/home/tommaso/scenicEx/data/matrix/checkpoints_overlap_250/train/model.ckpt-32670' #CHECKPOINT = '/home/tommaso/scenicEx/data/matrix/checkpoints/train/model.ckpt-5250' CHECKPOINT = '/home/tommaso/squeezeDet/data/webots/train/checkpoints/train/model.ckpt-1150' FILE = '/home/tommaso/squeezeDet/data/webots/test/images/1003.png' PREFIX = '/home/tommaso/scenicEx/data/matrix/' PREFIX_LABELS = PREFIX + 'labels/' PREFIX_IMAGES = PREFIX + 'images/' image_set = '/home/tommaso/scenicEx/data/matrix/ImageSets/test_overlap.txt' net = nn.init(CHECKPOINT) pred = nn.classify(FILE, net, NN_PROB_THRESH) print(pred) #ap, recall = eval_set(net, image_set) # avg_precs = [] # avg_recs = [] # max_precs = [] # max_recs = [] # # for i in range(1,9,1): # precs = [] # recs = [] # for j in range(4950,5020,10): # checkpoint_path = '/home/tommaso/scenicEx/data/matrix/checkpoints_overlap_250/checkpoints_overlap_250_' + str(i) + '/train/model.ckpt-' + str(j) # print(checkpoint_path)
'''Write predictions into files'''
f = open(preds_directory + filename + '.txt', "w")
for p in prediction:
kitti = box_2_kitti_format(p[2:])
f.write("Car " + str(p[1]) + " " + " ".join([str(x)
for x in kitti]) + '\n')
f.close()
print('saved to {}'.format(preds_directory + filename + '.txt'))
IOU_THRESH = 0.5
IOU2_THRESH = 0.7
NN_PROB_THRESH = 0.5
PREFIX = './data/data/training/'
PREFIX_LABELS = PREFIX + 'label/' # directory path to label
PREFIX_IMAGES = PREFIX + 'image/' # directory path to image
SAVE_DIRECTORY = 'model_10000_prediction_label/'
checkpoint = './data/model_checkpoints/squeezeDet/model.ckpt-10000'
# This .txt is generated by following instruction in squeezeDet repo README
# go to ImageSet and execute command :
# ls ../directory_path_to_image | grep ".png" | sed s/.png// > train.txt
# this command parses out only the names of images into a .txt file
test_set = './data/gta/ImageSets/testing2_3_correct_label.txt'
net = nn.init(checkpoint)
print("model initiated!!")
c = evaluate_set(net, test_set)
print(c)