Hi markshih, I first run your pretrained weights in 44 epoch then realize predict boxes are much bigger than objects.
I found they mentioned about prototxt file in Robert-JunWang/Pelee#9
What to be changed in you implementations?

Hi,

Could you please help me with this error?

I am using the same parameters as you:

model config

batch_size = 16
image_size = (300, 300, 3)
n_classes = 80
mode = 'training'
l2_regularization = 0.0005
min_scale = 0.1
max_scale = 0.9
scales = None
aspect_ratios_global = None
aspect_ratios_per_layer = [[1.0, 2.0, 0.5, 3.0, 1.0 / 3.0],
[1.0, 2.0, 0.5, 3.0, 1.0 / 3.0],
[1.0, 2.0, 0.5, 3.0, 1.0 / 3.0],
[1.0, 2.0, 0.5, 3.0, 1.0 / 3.0],
[1.0, 2.0, 0.5, 3.0, 1.0 / 3.0]]
two_boxes_for_ar1 = True
steps = None
offsets = None
clip_boxes = False
variances = [0.1, 0.1, 0.2, 0.2]
coords = 'centroids'
normalize_coords = True
subtract_mean = [123, 117, 104]
divide_by_stddev = 128
swap_channels = None
confidence_thresh = 0.01
iou_threshold = 0.45
top_k = 200
nms_max_output_size = 400
return_predictor_sizes = False
K.clear_session()

But I get the error related to 383 line in pelee.py:
predictions = Concatenate(axis=2, name='ssd_predictions')([cls, box, priorbox])

ValueError: A Concatenate layer requires inputs with matching shapes except for the concat axis. Got inputs shapes: [(None, 1182, 81), (None, 1182, 4), (None, 3888, 8)]