Deploy mobilenet-v1 on U200

[lloo099]

Hi @maltanar , as I already build mobilenet-v1 model for U200. What mean of ret[0/1] and why there is index error?

img_queue = setup_dataloader(val_dir, label_file, batch_size)

ok = 0
nok = 0
i = 0
while not img_queue.last_batch:
    imgs, lbls = img_queue.get_batch(batch_size, timeout=None)
    imgs = np.array(imgs)
    exp = np.array(lbls)
    
    ibuf_normal = imgs.reshape(accel.ishape_normal)
    obuf_normal = accel.execute(ibuf_normal)
    obuf_normal = obuf_normal.reshape(batch_size, -1)[:,0]
    ret = np.bincount(obuf_normal.flatten() == exp.flatten())
    nok += ret[0]
    ok += ret[1]
    i += 1
    print("batch %d : total OK %d NOK %d" % (i, ok, nok))

Index errbatch 1 : total OK 8 NOK 992

IndexError Traceback (most recent call last)
in
14 ret = np.bincount(obuf_normal.flatten() == exp.flatten())
15 nok += ret[0]
---> 16 ok += ret[1]
17 i += 1
18 print("batch %d : total OK %d NOK %d" % (i, ok, nok))

IndexError: index 1 is out of bounds for axis 0 with size 1or:

[lloo099]

@fpjentzsch When I change nok += ret[0] and ok += ret[1] to nok += ret[0] and ok += ret[0] -> Final top-1 accuracy: 50.0%
nok += ret[-1] and ok += ret[0] -> Final top-1 accuracy: 52.09473684210526%
nok += ret[0] and ok += ret[-1] -> Final top-1 accuracy: 47.915789473684214%

Why this happened?
my dataset: Small-ImageNet-Validation-Dataset-1000-Classes(https://github.com/ndb796/Small-ImageNet-Validation-Dataset-1000-Classes)