def _evaluate(Gt, Hy, threshold, cost_fun, info=False, debug_info=False):
if debug_info:
fp, m, mme, c, d, g, FN_pairs, FP_pairs, MME_pairs = MOTM.evaluate(
Gt, Hy, threshold, cost_fun, debug_info=True)
else:
fp, m, mme, c, d, g = MOTM.evaluate(Gt, Hy, threshold, cost_fun)
FN = np.sum(m)
FP = np.sum(fp)
IDSW = np.sum(mme)
GT = np.sum(g)
mota = 1 - (FN + FP + IDSW) / GT
result = [mota]
if info:
result.append({'FN': FN, 'FP': FP, 'IDSW': IDSW, 'GT': GT})
if debug_info:
result.append({"FN": FN_pairs, "FP": FP_pairs, "IDSW": MME_pairs})
if len(result) > 1:
return result
else:
return result[0]
def test_motm_1mme(self):
Gt = np.array([
[1, 1, 0, 0],
[2, 1, 0, 0]
])
Hy = np.array([
[1, 1, 0, 0],
[2, 2, 0, 0]
])
cost = lambda a, b: la.norm(a-b)
fp, m, mme, c, d, g = MOTM.evaluate(Gt, Hy, 10, cost)
self.assertEqual(len(fp), 2)
self.assertEqual(len(m), 2)
self.assertEqual(len(mme), 2)
self.assertEqual(len(c), 2)
self.assertEqual(len(d), 2)
self.assertEqual(len(g), 2)
self.assertEqual(np.sum(fp), 0)
self.assertEqual(np.sum(m), 0)
self.assertEqual(np.sum(mme), 1)
self.assertEqual(np.sum(d), 0)
self.assertEqual(np.sum(g), 2)
self.assertEqual(np.sum(c), 2)
def test_delete(self):
Data = np.array([
[1, 1, None],
[1, 2, None],
[1, 3, None],
[1, 4, None],
])
data = MOTM.SingleFrameData(Data)
self.assertEqual(data.total_elements, 4)
self.assertEqual(data.elements_left, 4)
data.remove([1, None])
self.assertEqual(data.total_elements, 4)
self.assertEqual(data.elements_left, 3)
data.remove([4, None])
self.assertEqual(data.total_elements, 4)
self.assertEqual(data.elements_left, 2)
data.remove([3, None])
self.assertEqual(data.total_elements, 4)
self.assertEqual(data.elements_left, 1)
data.remove([2, None])
self.assertEqual(data.total_elements, 4)
self.assertEqual(data.elements_left, 0)
self.assertEqual(len(data.as_list()), 0)
def test_basics(self):
Data = np.array([(1, 1), (1, 2), (1, 3), (1, 4), (2, 1), (2, 2),
(2, 3), (3, 1), (3, 2)])
L = MOTM.LocalFrameIdToGlobal(Data)
self.assertEqual(L.get_true_idx(1, 0), 0)
self.assertEqual(L.get_true_idx(1, 2), 2)
self.assertEqual(L.get_true_idx(3, 1), 8)
def test_out_of_order(self):
Data = np.array([(2, 3), (1, 2), (3, 1), (1, 4), (1, 3), (2, 2),
(1, 1), (3, 2), (2, 1)])
L = MOTM.LocalFrameIdToGlobal(Data)
self.assertEqual(L.get_true_idx(2, 0), 0)
self.assertEqual(L.get_true_idx(1, 2), 4)
self.assertEqual(L.get_true_idx(3, 1), 7)
def test_empty(self):
Data = np.array([])
data = MOTM.SingleFrameData(Data)
self.assertEqual(data.total_elements, 0)
self.assertEqual(data.elements_left, 0)
A = data.as_list()
self.assertEqual(len(A), 0)
def test_mismatch_inverse(self):
M = MOTM.MatchLookup(1, 2)
o, h = (1, None), (99, None)
M.insert_match(1, o, h)
o, h = (2, None), (99, None)
M.insert_match(2, o, h)
self.assertFalse(M.has_mismatch(1, o, h))
self.assertTrue(M.has_mismatch(2, o, h))
self.assertEqual(M.count_matches(1), 1)
self.assertEqual(M.count_matches(2), 1)
def test_setup(self):
Data = np.array([
[1, 1, None],
[1, 2, None],
[1, 3, None],
[1, 4, None],
])
data = MOTM.SingleFrameData(Data)
self.assertEqual(data.total_elements, 4)
self.assertEqual(data.elements_left, 4)
def test_complex_find(self):
cost_fun = lambda a, b: abs(a - b)
Data = np.array([
[1, 1, 10],
[1, 2, 12],
[1, 2, 100],
[1, 2, -19],
])
data = MOTM.SingleFrameData(Data)
self.assertEqual(data.total_elements, 4)
self.assertEqual(data.elements_left, 4)
obj = data.find_best(2, [1, 10], cost_fun)
self.assertEqual(obj[0], 2)
self.assertEqual(obj[1], 12)
def test_listing(self):
Data = np.array([
[1, 1, None],
[1, 2, None],
[1, 3, None],
[1, 4, None],
])
data = MOTM.SingleFrameData(Data)
data.remove([3, None])
data.remove([2, None])
A = data.as_list()
self.assertEqual(len(A), 2)
self.assertEqual(A[0][0], 1)
self.assertEqual(A[1][0], 4)
def test_motm_fp_with_debug_info(self):
Gt = np.array([
[1, 1, 0, 0],
[1, 2, 10, 10],
[2, 1, 0, 0]
])
Hy = np.array([
[1, 2, 10, 10],
[1, 3, 20, 20],
[1, 4, 30, 30],
[2, 1, 0, 0],
[1, 1, 0, 0],
[2, 5, 88, 99]
])
cost = lambda a, b: la.norm(a-b)
fp, m, mme, c, d, g, FN_pairs, FP_pairs, MME_pairs =\
MOTM.evaluate(Gt, Hy, 5, cost,
debug_info=True)
self.assertEqual(len(FN_pairs), 0)
self.assertEqual(len(FP_pairs), 3)
self.assertEqual(len(MME_pairs), 0)
self.assertEqual(FP_pairs[0][0], 1)
self.assertEqual(FP_pairs[1][0], 1)
self.assertEqual(FP_pairs[2][0], 2)
self.assertEqual(FP_pairs[0][2], 20)
self.assertEqual(FP_pairs[1][2], 30)
self.assertEqual(FP_pairs[2][2], 88)
self.assertEqual(len(fp), 2)
self.assertEqual(len(m), 2)
self.assertEqual(len(mme), 2)
self.assertEqual(len(c), 2)
self.assertEqual(len(d), 2)
self.assertEqual(len(g), 2)
self.assertEqual(np.sum(fp), 3)
self.assertEqual(np.sum(m), 0)
self.assertEqual(np.sum(mme), 0)
self.assertEqual(np.sum(d), 0)
self.assertEqual(np.sum(g), 3)
self.assertEqual(np.sum(c), 3)
def test_multiple_inserts(self):
M = MOTM.MatchLookup(1, 2)
o, h = (1, None), (95.5, None)
M.insert_match(1, o, h)
o, h = (2, None), (96, None)
M.insert_match(1, o, h)
o, h = (3, None), (97, None)
M.insert_match(1, o, h)
self.assertEqual(len(M.get_matches(1)), 3)
o, h = (2, None), (96, None)
M.insert_match(2, o, h)
o, h = (3, None), (97, None)
M.insert_match(2, o, h)
self.assertEqual(len(M.get_matches(2)), 2)
def test_counting(self):
M = MOTM.MatchLookup(1, 2)
o, h = (1, None), (95, None)
M.insert_match(1, o, h)
o, h = (2, None), (96, None)
M.insert_match(1, o, h)
o, h = (3, None), (97, None)
M.insert_match(1, o, h)
o, h = (3, None), (97, None)
M.insert_match(2, o, h)
self.assertEqual(M.count_matches(1), 3)
self.assertEqual(M.count_matches(2), 1)
def test_insert(self):
M = MOTM.MatchLookup(1, 2)
self.assertEqual(len(M.get_matches(0)), 0)
o, h = (1, None), (99, None)
M.insert_match(1, o, h)
self.assertEqual(len(M.get_matches(1)), 1)
self.assertFalse(M.has_mismatch(1, o, h))
o_, h_ = M.get_matches(1)[0]
self.assertEqual(o_[0], 1)
self.assertEqual(h_[0], 99)
o, h = (1, None), (99, None)
M.insert_match(2, o, h)
self.assertFalse(M.has_mismatch(2, o, h))
def evaluate(Gt, Hy, threshold):
""" Ground-truth vs hypothesis for the
Multiple Object Tracking Precision
Gt: [
(frame, pid, x, y)
]
Hy: [
(frame, pid, x, y)
]
threshold: after which no correspondence is possible
"""
cost_fun = lambda a, b: la.norm(a-b)
fp, m, mme, c, d, g = MOTM.evaluate(Gt, Hy, threshold, cost_fun)
D = np.sum(d)
C = np.sum(c)
return D/C
def test_multiple_delete(self):
Data = np.array([[1, 1, 99], [1, 2, 88], [1, 2, 99], [1, 3, 77],
[1, 2, 77]])
data = MOTM.SingleFrameData(Data)
self.assertEqual(data.total_elements, 5)
self.assertEqual(data.elements_left, 5)
data.remove([2, 99])
self.assertEqual(data.elements_left, 4)
data.remove([1, 99])
self.assertEqual(data.elements_left, 3)
A = data.as_list()
self.assertEqual(len(A), 3)
self.assertEqual(A[0][0], 2)
self.assertEqual(A[0][1], 88)
self.assertEqual(A[1][0], 2)
self.assertEqual(A[1][1], 77)
self.assertEqual(A[2][0], 3)
self.assertEqual(A[2][1], 77)
def test_find(self):
Data = np.array([
[1, 1, 66],
[1, 2, 77],
[1, 3, 88],
[1, 4, 99],
])
data = MOTM.SingleFrameData(Data)
self.assertEqual(data.total_elements, 4)
self.assertEqual(data.elements_left, 4)
obj = data.find(1)
self.assertEqual(obj[0], 1)
self.assertEqual(obj[1], 66)
obj = data.find(2)
self.assertEqual(obj[0], 2)
self.assertEqual(obj[1], 77)
obj = data.find(88)
self.assertEqual(obj, None)
def test_motm_interrupt(self):
Gt = np.array([
[1, 1, 0, 0],
[2, 1, 0, 0],
[5, 1, 0, 0],
[6, 1, 0, 0],
[7, 1, 0, 0],
[8, 1, 0, 0]
])
Hy = np.array([
[1, 1, 0, 0],
[2, 1, 0, 0],
[5, 1, 0, 0],
[6, 1, 0, 0],
[7, 1, 0, 0],
[8, 1, 0, 0]
])
cost = lambda a, b: la.norm(a-b)
fp, m, mme, c, d, g = MOTM.evaluate(Gt, Hy, 10, cost)
self.assertEqual(len(fp), 8)
self.assertEqual(len(m), 8)
self.assertEqual(len(mme), 8)
self.assertEqual(len(c), 8)
self.assertEqual(len(d), 8)
self.assertEqual(len(g), 8)
self.assertEqual(np.sum(fp), 0)
self.assertEqual(np.sum(m), 0)
self.assertEqual(np.sum(mme), 0)
self.assertEqual(np.sum(d), 0)
self.assertEqual(np.sum(g), 6)
self.assertEqual(np.sum(c), 6)
