Is some BUG in bbox_transform.py??

[tanghy2016]

I added another class to the old 20 ones, but after that, when the training iterated about 20,000 times, overflow occurred and the program ended. So I wrote a print sentence in right that function which overflowed before for checking, and below are the results. Do u know anything about my problem?

code as follow:

def bbox_transform_inv(boxes, deltas): if boxes.shape[0] == 0: return np.zeros((0, deltas.shape[1]), dtype=deltas.dtype)

boxes = boxes.astype(deltas.dtype, copy=False)

widths = boxes[:, 2] - boxes[:, 0] + 1.0
heights = boxes[:, 3] - boxes[:, 1] + 1.0
ctr_x = boxes[:, 0] + 0.5 * widths
ctr_y = boxes[:, 1] + 0.5 * heights

dx = deltas[:, 0::4]
dy = deltas[:, 1::4]
dw = deltas[:, 2::4]
dh = deltas[:, 3::4]

print 'dx=\n', dx, 'dy=\n', dy, 'dw=\n', dw, 'dh=\n', dh, 'shape', dx.shape
pred_ctr_x = dx * widths[:, np.newaxis] + ctr_x[:, np.newaxis]
pred_ctr_y = dy * heights[:, np.newaxis] + ctr_y[:, np.newaxis]
pred_w = np.exp(dw) * widths[:, np.newaxis]
pred_h = np.exp(dh) * heights[:, np.newaxis]

pred_boxes = np.zeros(deltas.shape, dtype=deltas.dtype)
# x1
pred_boxes[:, 0::4] = pred_ctr_x - 0.5 * pred_w
# y1
pred_boxes[:, 1::4] = pred_ctr_y - 0.5 * pred_h
# x2
pred_boxes[:, 2::4] = pred_ctr_x + 0.5 * pred_w
# y2
pred_boxes[:, 3::4] = pred_ctr_y + 0.5 * pred_h

return pred_boxes

result as follow:

• I1107 20:13:32.075683 1792 solver.cpp:229] Iteration 16580, loss = nan I1107 20:13:32.075737 1792 solver.cpp:245] Train net output #0: loss_bbox = 0 ( 1 = 0 loss) I1107 20:13:32.075749 1792 solver.cpp:245] Train net output #1: loss_cls = 0.235863 ( 1 = 0.235863 loss) I1107 20:13:32.075757 1792 solver.cpp:245] Train net output #2: rpn_cls_los s = 2.20898 ( 1 = 2.20898 loss) I1107 20:13:32.075765 1792 solver.cpp:245] Train net output #3: rpn_loss_bb ox = nan ( 1 = nan loss) I1107 20:13:32.075773 1792 sgd_solver.cpp:106] Iteration 16580, lr = 0.001 dx= [[ 1.69262147] [-0.50489259] [ 0.30272573] ..., [ 1.80490637] [ 0.45227113] [ 0.64777136]] dy= [[ 15.27661228] [ 0.41482714] [ 0.06253394] ..., [ 15.32104683] [ 1.0138334 ] [ 0.82928151]] dw= [[ 0.86533314] [ 0.12271755] [ 0.06882791] ..., [ 1.4919914 ] [ 0.10998501] [ 0.46326059]] dh= [[ 1.89869058] [ 0.04458962] [-0.15501517] ..., [ 0.95262975] [-0.11793157] [-0.65379727]] shape (19494, 1) dx= [[ 1.51812971] [-0.44130734] [ 0.21798934] ..., [ 1.50871968] [ 0.30029866] [ 0.5220989 ]] dy= [[ 13.10294056] [ 0.37086505] [ 0.0634751 ] ..., [ 13.35043907] [ 0.80506349] [ 0.71783197]] dw= [[ 0.81970924] [ 0.1567906 ] [ 0.06341352] ..., [ 1.35228169] [ 0.07805473] [ 0.42007568]] dh= [[ 1.65305829] [ 0.05194709] [-0.22186428] ..., [ 1.07120883] [-0.00576194] [-0.5921734 ]] shape (17100, 1) dx= [[ 1.93541121] [-0.60172379] [ 0.39276949] ..., [ 1.0219456 ] [ 0.22475462] [ 0.38502726]] dy= [[ 18.6795311 ] [ 0.51199323] [ 0.09984528] ..., [ 8.81377983] [ 0.57531947] [ 0.46192282]] dw= [[ 0.91600668] [ 0.09472741] [ 0.05817838] ..., [ 0.8552379 ] [ 0.10623112] [ 0.31314957]] dh= [[ 2.05678964] [ 0.00360752] [-0.209602 ] ..., [ 0.54301333] [ 0.00342666] [-0.38243169]] shape (17442, 1) dx= [[ 0.69756079] [-0.20435131] [ 0.05022508] ..., [ 2.20087218] [ 0.5320493 ] [ 0.87700069]] dy= [[ 5.28811502] [ 0.13694541] [ 0.04660323] ..., [ 20.7341938 ] [ 1.2682693 ] [ 1.09226692]] dw= [[ 0.52218628] [ 0.14009777] [ 0.08773904] ..., [ 1.70942426] [ 0.07307827] [ 0.68221164]] dh= [[ 0.90885699] [ 0.12151945] [-0.01091111] ..., [ 1.27402627] [-0.21292779] [-0.86571634]] shape (17784, 1) dx= [[ 14.12853718] [ -4.58679676] [ 3.86513996] ..., [ 14.44054508] [ 4.16536045] [ 6.94221354]] dy= [[ 153.39071655] [ 4.67038727] [ 0.44393685] ..., [ 157.16017151] [ 9.22263908] [ 8.33644772]] dw= [[ 4.85911083] [ 0.28020942] [-0.41960421] ..., [ 9.76607418] [-0.43815124] [ 5.00803804]] dh= [[ 14.52654457] [ -1.0439378 ] [ -1.8397969 ] ..., [ 7.91534948] [ -2.18607712] [ -6.84572887]] shape (19494, 1) /deep/py-faster-rcnn/tools/../lib/fast_rcnn/bbox_transform.py:49: RuntimeWarning : overflow encountered in exp pred_w = np.exp(dw) widths[:, np.newaxis] /deep/py-faster-rcnn/tools/../lib/fast_rcnn/bbox_transform.py:49: RuntimeWarning : overflow encountered in multiply pred_w = np.exp(dw) widths[:, np.newaxis] dx= [[ 17.34057236] [ -5.66221857] [ 4.7488308 ] ..., [ 21.51120758] [ 6.1817255 ] [ 10.1119833 ]] dy= [[ 189.39860535] [ 5.8252492 ] [ 0.64366901] ..., [ 230.1686554 ] [ 13.73810291] [ 12.21232796]] dw= [[ 5.86071014] [ 0.24205765] [ -0.50441903] ..., [ 14.99213219] [ -0.56799537] [ 7.20289278]] dh= [[ 17.81727219] [ -1.33075023] [ -2.29880214] ..., [ 11.93072891] [ -3.11075139] [-10.10451603]] shape (17100, 1) dx= [[ 0.1809919 ] [ 0.05314905] [-0.00585624] ..., [ 0.36789721] [ 0.07726575] [ 0.17214988]] dy= [[ 9.60180521e-01] [ -3.28403860e-02] [ -3.10134888e-03] ..., [ 4.03634357e+00] [ 2.36444727e-01] [ 2.12220207e-01]] dw= [[ 0.25702038] [ 0.2446238 ] [ 0.08647276] ..., [ 0.17067978] [ 0.04406154] [ 0.14802027]] dh= [[ 0.32531762] [ 0.20668527] [ 0.09408868] ..., [ 0.01454777] [ 0.01304469] [-0.13869783]] shape (19494, 1) dx= [[ 0.32107556] [ 0.03948319] [ 0.00929752] ..., [ 1.21000493] [ 0.20505081] [ 0.5341391 ]] dy= [[ 1.32112694] [ -0.06508708] [ -0.03040531] ..., [ 13.55274487] [ 0.70385402] [ 0.70753902]] dw= [[ 0.48596889] [ 0.32714739] [ 0.11804374] ..., [ 0.73939699] [ 0.00370674] [ 0.41292527]] dh= [[ 0.67309242] [ 0.28599161] [ 0.07322749] ..., [ 0.42121381] [-0.10930666] [-0.48818183]] shape (17100, 1) dx= [[ 0.28732941] [ 0.16537723] [-0.00154071] ..., [ 0.34561181] [ 0.01218602] [ 0.09984097]] dy= [[ 0.95636111] [-0.14533311] [-0.05571485] ..., [ 1.79893255] [ 0.08993389] [ 0.05374222]] dw= [[ 0.37399614] [ 0.29909372] [ 0.07877787] ..., [ 0.21127498] [ 0.09000482] [ 0.0605586 ]] dh= [[ 0.68116724] [ 0.37220293] [ 0.17048553] ..., [ 0.10395312] [ 0.16595674] [-0.05677746]] shape (17100, 1) dx= [[ 0.25039169] [ 0.07315145] [-0.00675968] ..., [ 0.33019078] [ 0.03063683] [ 0.06724098]] dy= [[ 0.7646212 ] [-0.1379102 ] [-0.03177064] ..., [ 1.82896745] [ 0.14514422] [ 0.08653554]] dw= [[ 0.41362 ] [ 0.34356815] [ 0.07996602] ..., [ 0.32493946] [ 0.10698123] [ 0.07901177]] dh= [[ 0.70002955] [ 0.37839428] [ 0.17210624] ..., [ 0.11815372] [ 0.10741498] [-0.06781152]] shape (17100, 1) dx= [[ 0.20503215] [ 0.14187868] [-0.02538324] ..., [ 0.25761825] [ 0.00063007] [ 0.03702011]] dy= [[ 0.73968458] [-0.2088578 ] [-0.03802263] ..., [ 1.08071482] [ 0.10128661] [ 0.01579206]] dw= [[ 0.37786177] [ 0.39688873] [ 0.09559654] ..., [ 0.37650517] [ 0.11314878] [ 0.00104013]] dh= [[ 0.66281664] [ 0.36328161] [ 0.12070411] ..., [ 0.09586006] [ 0.12889241] [-0.07499333]] shape (17100, 1) dx= [[ 0.21474786] [ 0.21825135] [-0.03084789] ..., [ 0.18840194] [-0.02748936] [ 0.00671676]] dy= [[ 0.70849383] [-0.2030239 ] [-0.05798998] ..., [ 0.46467435] [-0.00881376] [-0.03415608]] dw= [[ 0.32540816] [ 0.38482958] [ 0.06325729] ..., [ 0.21186957] [ 0.0916909 ] [-0.01272086]] dh= [[ 0.68713808] [ 0.39726567] [ 0.19117579] ..., [ 0.06136551] [ 0.16881415] [-0.05074813]] shape (19494, 1) dx= [[ 0.22441402] [ 0.36710414] [-0.03080902] ..., [ 0.1880618 ] [-0.0090771 ] [ 0.02065454]] dy= [[ 0.91155028] [-0.39651006] [-0.10748966] ..., [ 0.613711 ] [ 0.03552385] [-0.04408681]] dw= [[ 0.44617409] [ 0.50943768] [ 0.0759024 ] ..., [ 0.08976659] [ 0.05554071] [-0.00567405]] dh= [[ 0.90590668] [ 0.56420249] [ 0.24491608] ..., [-0.00752416] [ 0.1331844 ] [-0.06337131]] shape (19494, 1) dx= [[ 0.26826978] [ 0.33719975] [ 0.00888054] ..., [-0.00069787] [ 0.00183878] [-0.01399136]] dy= [[ 0.79242063] [-0.2804276 ] [-0.10856268] ..., [ 0.00172816] [-0.01159885] [-0.07661776]] dw= [[ 0.46090895] [ 0.55589867] [ 0.03495556] ..., [-0.05373206] [ 0.05413384] [-0.10359898]] dh= [[ 0.97432417] [ 0.5159874 ] [ 0.31585801] ..., [-0.23290846] [ 0.05915573] [-0.04164039]] shape (14175, 1) dx= [[ 0.10047725] [ 0.20022117] [-0.03589112] ..., [-0.08821255] [-0.01690185] [-0.00311243]] dy= [[ 0.38369769] [-0.18052378] [-0.02655566] ..., [-0.1624622 ] [-0.13792492] [-0.14525057]] dw= [[ 0.21880329] [ 0.31190559] [ 0.07283732] ..., [-0.52908111] [-0.10553086] [-0.14146115]] dh= [[ 0.44032821] [ 0.31888157] [ 0.18683948] ..., [-0.40582103] [ 0.04411977] [-0.07514375]] shape (12996, 1) dx= [[ 0.1109004 ] [ 0.28959665] [-0.0285324 ] ..., [-0.16186452] [ 0.04103525] [ 0.03899656]] dy= [[ 0.40375963] [-0.25693163] [-0.06114756] ..., [-0.26582715] [-0.2584942 ] [-0.32886657]] dw= [[ 0.21400028] [ 0.38730782] [ 0.02163425] ..., [-0.82393658] [-0.17964846] [-0.31661227]] dh= [[ 0.55946749] [ 0.43363559] [ 0.23507506] ..., [-0.48474723] [ 0.00101788] [-0.15810674]] shape (19494, 1) dx= [[ 0.41041312] [ 0.9134481 ] [-0.07639228] ..., [-0.13172315] [ 0.00921583] [-0.09478085]] dy= [[ 1.72121632] [-0.85182846] [-0.22939041] ..., [ 0.06709117] [-0.04727058] [-0.12477416]] dw= [[ 0.41443938] [ 0.99802607] [ 0.20317978] ..., [-0.06451301] [ 0.10018457] [-0.11905614]] dh= [[ 0.91670465] [ 0.97788846] [ 0.29120368] ..., [-0.38113612] [-0.10865723] [-0.00260712]] shape (17100, 1) dx= [[ 0.45247287] [ 0.81061602] [-0.0817315 ] ..., [-0.18486582] [ 0.02051136] [-0.10640365]] dy= [[ 1.64135408] [-0.77393419] [-0.18273427] ..., [-0.06820877] [-0.04259291] [-0.127818 ]] dw= [[ 0.46278149] [ 0.97440398] [ 0.20806728] ..., [-0.00675257] [ 0.09623937] [-0.15885375]] dh= [[ 0.9344781 ] [ 0.9387536 ] [ 0.27473497] ..., [-0.46198261] [-0.23432632] [ 0.03853504]] shape (17100, 1) dx= [[ 1.03430223] [ 2.30005026] [-0.11117247] ..., [-0.02603048] [-0.05089114] [-0.04210172]] dy= [[ 4.20332146] [-2.10101485] [-0.44300577] ..., [ 0.47191751] [ 0.02904183] [-0.01401345]] dw= [[ 0.73855132] [ 2.27713919] [ 0.43637192] ..., [-0.12361873] [ 0.17306626] [-0.07348016]] dh= [[ 1.90010715] [ 2.51771855] [ 0.88009691] ..., [-0.55456424] [ 0.03669987] [-0.0603053 ]] shape (17100, 1) dx= [[ 0.20431782] [ 0.4666419 ] [-0.04235181] ..., [-0.04195286] [-0.02634729] [-0.08359165]] dy= [[ 0.9552567 ] [-0.39553213] [-0.0751732 ] ..., [ 0.19391023] [-0.02668684] [-0.04990452]] dw= [[ 0.20384723] [ 0.53485441] [ 0.13949443] ..., [ 0.02948399] [ 0.1293415 ] [-0.0338269 ]] dh= [[ 0.41977543] [ 0.60789013] [ 0.23389585] ..., [-0.3617245 ] [ 0.03751442] [-0.0058842 ]] shape (17100, 1) dx= [[ 1.35073531] [ 5.67767382] [-0.24248599] ..., [ 0.78988332] [ 0.25262311] [-0.49636576]] dy= [[ 11.11100006] [ -5.11032581] [ -1.27450335] ..., [ 3.31481028] [ -0.24454108] [ 0.52495456]] dw= [[ 1.87556434] [ 6.77083302] [ 0.70132756] ..., [-0.59234947] [ 0.7528131 ] [ 0.2544432 ]] dh= [[ 3.51630402] [ 5.8298521 ] [ 1.23896706] ..., [-1.17637324] [ 1.25345218] [-0.35519907]] shape (17100, 1) dx= [[ 0.96406072] [ 3.78742957] [-0.11582164] ..., [ 0.23537992] [ 0.12082122] [-0.19350524]] dy= [[ 7.66653156] [-3.34219956] [-0.7990135 ] ..., [ 1.2024318 ] [-0.14002384] [ 0.16317572]] dw= [[ 1.14247572] [ 4.42902613] [ 0.43920061] ..., [-0.45944577] [ 0.38106596] [ 0.24805646]] dh= [[ 2.42970657] [ 4.10410357] [ 0.8594563 ] ..., [-0.62541968] [ 0.60564679] [-0.20369534]] shape (17100, 1) dx= [[ 0.10697325] [ 0.43582934] [-0.05002928] ..., [ 0.0515424 ] [-0.08455836] [ 0.04180375]] dy= [[ 0.90025508] [-0.40443635] [-0.07439174] ..., [ 0.8137722 ] [ 0.0908379 ] [ 0.07157324]] dw= [[ 0.17972851] [ 0.53842109] [ 0.13666055] ..., [ 0.1901798 ] [ 0.12636837] [-0.02954175]] dh= [[ 0.25491041] [ 0.55249894] [ 0.21506912] ..., [-0.10795481] [ 0.08906633] [-0.03510559]] shape (17442, 1) dx= [[ 0.22961064] [ 1.22430336] [-0.09720994] ..., [ 0.04951505] [-0.14887942] [ 0.05519611]] dy= [[ 2.28245068] [-1.17654264] [-0.23033002] ..., [ 1.18968403] [ 0.10551875] [ 0.10819419]] dw= [[ 0.26413679] [ 1.32965767] [ 0.26111427] ..., [ 0.35542899] [ 0.14100833] [-0.07186134]] dh= [[ 0.5100441 ] [ 1.32613301] [ 0.4117521 ] ..., [-0.02236348] [ 0.10099887] [-0.06326412]] shape (16416, 1) dx= [[ 0.46102053] [ 2.76373315] [-0.50167447] ..., [ 0.49863881] [-0.45490998] [ 0.86568892]] dy= [[ 4.28244543] [-2.78910422] [-0.24722575] ..., [ 2.73887539] [-0.14832316] [ 0.18437037]] dw= [[ 0.68512988] [ 2.91124272] [ 0.63992149] ..., [ 0.39605355] [-0.47715092] [-0.89671916]] dh= [[ 0.96098715] [ 2.59232569] [ 0.3644039 ] ..., [-1.35580564] [ 0.36125544] [-1.28961217]] shape (18810, 1) dx= [[ 0.38278148] [ 2.21001506] [-0.40704727] ..., [ 0.21528383] [-0.22215353] [ 0.41980946]] dy= [[ 3.44531107] [-2.22523761] [-0.19572181] ..., [ 1.33708417] [-0.05370709] [ 0.07799486]] dw= [[ 0.57630223] [ 2.35486388] [ 0.52876341] ..., [ 0.13793305] [-0.16817886] [-0.38649681]] dh= [[ 0.79142112] [ 2.10395479] [ 0.31321391] ..., [-0.70205843] [ 0.21892643] [-0.59913343]] shape (17100, 1) dx= [[ 2.04116869] [ 14.38851452] [ -1.93457472] ..., [ 1.40617251] [ 0.84620243] [ 1.50956523]] dy= [[ 22.12321663] [-13.18910885] [ -2.07902503] ..., [ 37.72657013] [ 2.71346188] [ 1.76873934]] dw= [[ 8.51811409e-01] [ 1.51878347e+01] [ 2.10579228e+00] ..., [ 1.38966814e-02] [ 6.76388144e-01] [ 1.21595144e+00]] dh= [[ 2.2326889 ] [ 13.98908997] [ 1.44217658] ..., [ -0.4036755 ] [ -0.4538337 ] [ -1.56191313]] shape (19152, 1) dx= [[ 4.75547504] [ 32.46413803] [ -4.46559191] ..., [ 1.31493771] [ 1.08607757] [ 1.94233704]] dy= [[ 49.30391312] [-29.86575317] [ -4.61424875] ..., [ 46.36717987] [ 3.31236601] [ 2.1841588 ]] dw= [[ 1.94153488] [ 34.15163422] [ 4.75500631] ..., [ -0.72264206] [ 0.712551 ] [ 1.54035687]] dh= [[ 4.97435951] [ 31.34772682] [ 3.0725832 ] ..., [ -0.7400651 ] [ -0.5795961 ] [ -1.96256447]] shape (17100, 1) dx= [[ 128079.4609375 ] [ 897339.0625 ] [ 431199.0625 ] ..., [-185503.484375 ] [-145269.9375 ] [ 2974.33032227]] dy= [[ 922321. ] [-891607.875 ] [ 603538.75 ] ..., [ 176326.921875 ] [ -70948.765625 ] [-116401.3984375]] dw= [[ 217711.3125 ] [ 733906.8125 ] [-396041.65625 ] ..., [ 197423.125 ] [-234642.484375] [ -95428.859375]] dh= [[ 1203753. ] [ 1177189.75 ] [ -539389.125 ] ..., [ -101695.984375] [ -252951.625 ] [ -642465.4375 ]] shape (14022, 1) dx= [[ 322652.21875 ] [ 2389257.25 ] [ 1140256.125 ] ..., [ -175917.78125 ] [ -136041.171875 ] [ 4092.6081543]] dy= [[ 2471479.5 ] [-2371688.75 ] [ 1567701.625 ] ..., [ 166866.171875 ] [ -63144.83203125] [ -110667.6484375 ]] dw= [[ 592151.1875 ] [ 1984461.125 ] [-1052206.25 ] ..., [ 178742.359375] [ -221431.25 ] [ -88184.109375]] dh= [[ 3180583.25 ] [ 3128019.75 ] [-1423802.75 ] ..., [ -94561.1484375] [ -231314.109375 ] [ -596328.125 ]] shape (17100, 1) dx= [[ -1.99462084e+38] [ -6.97799265e+37] [ 7.32326951e+32] ..., [ 3.66380281e+37] [ 3.02497080e+37] [ 6.05244988e+31]] dy= [[ -1.88442752e+38] [ 6.97793789e+37] [ 9.39286927e+32] ..., [ 1.04679050e+37] [ 3.02497003e+37] [ -6.88375708e+31]] dw= [[ -7.08219926e+32] [ -6.74972570e+32] [ -2.59587496e+32] ..., [ -1.65612934e+32] [ -5.85255158e+31] [ -7.71560056e+31]] dh= [[ -1.99462672e+38] [ -6.97802967e+37] [ 4.62670660e+31] ..., [ 3.66377999e+37] [ 3.02498144e+37] [ -1.89055746e+32]] shape (19494, 1) /deep/py-faster-rcnn/tools/../lib/fast_rcnn/bbox_transform.py:47: RuntimeWarning: overflow encountered in multiply pred_ctr_x = dx widths[:, np.newaxis] + ctr_x[:, np.newaxis] /deep/py-faster-rcnn/tools/../lib/fast_rcnn/bbox_transform.py:48: RuntimeWarning: overflow encountered in multiply pred_ctr_y = dy heights[:, np.newaxis] + ctr_y[:, np.newaxis] /deep/py-faster-rcnn/tools/../lib/fast_rcnn/bbox_transform.py:56: RuntimeWarning: invalid value encountered in subtract pred_boxes[:, 1::4] = pred_ctr_y - 0.5 pred_h /deep/py-faster-rcnn/tools/../lib/fast_rcnn/bbox_transform.py:58: RuntimeWarning: invalid value encountered in add pred_boxes[:, 2::4] = pred_ctr_x + 0.5 pred_w /deep/py-faster-rcnn/tools/../lib/fast_rcnn/bbox_transform.py:60: RuntimeWarning: invalid value encountered in add pred_boxes[:, 3::4] = pred_ctr_y + 0.5 * pred_h /deep/py-faster-rcnn/tools/../lib/rpn/proposal_layer.py:175: RuntimeWarning: invalid value encountered in greater_equal keep = np.where((ws >= min_size) & (hs >= min_size))[0]

[wangsky2001]

I have a similar problem too. Any one find a solution?

[fernandorovai]

I have the same problem as well. Did you guys find any clue?

[HTLife]

I got similar problem. Didn't find any solution yet.

Every time when program goes to iter=55

gt_heights in bbox_transform will become negative.

def bbox_transform(ex_rois, gt_rois):
...
print(gt_rois[1, 3])
print(gt_rois[1, 1])

#normal
187.5
22.5
166.0

#error
188.75
81918.8
-81729.0

And it is caused by gt_rois[:, 1] is greater than gt_rois[:, 3]

gt_heights = gt_rois[:, 3] - gt_rois[:, 1] + 1.0

[HaiyongJiang]

I think it is because of the bug in the bbox_inv_transform. When we need to inversely calculate the x_right, y_top, we need to minus 1 for consistency, namely pred_boxes[:, 2::4] = pred_ctr_x + 0.5 * pred_w -1.

[wubaorong]

@tanghy2016 I met the same error,but I don't know how to solve it,can you help me? /home/wu/faster_rcnn/py-faster-rcnn/tools/../lib/fast_rcnn/bbox_transform.py:51: RuntimeWarning: overflow encountered in exp pred_w = np.exp(dw) widths[:, np.newaxis] /home/wu/faster_rcnn/py-faster-rcnn/tools/../lib/fast_rcnn/bbox_transform.py:51: RuntimeWarning: overflow encountered in multiply pred_w = np.exp(dw) widths[:, np.newaxis] /home/wu/faster_rcnn/py-faster-rcnn/tools/../lib/fast_rcnn/bbox_transform.py:52: RuntimeWarning: overflow encountered in exp pred_h = np.exp(dh) heights[:, np.newaxis] /home/wu/faster_rcnn/py-faster-rcnn/tools/../lib/fast_rcnn/bbox_transform.py:52: RuntimeWarning: overflow encountered in multiply pred_h = np.exp(dh) heights[:, np.newaxis] dw [[ -43.27822113] [ -25.07061386] [ 2.01593685] ..., [-656.98260498] [ -1.75763202] [ 8.05360603]] dh [[ -15.3271513 ] [ -91.23004913] [ 10.69660854] ..., [-882.39678955] [ 5.89947605] [ -46.34132004]] dw [[ 3.35388470e+00] [ 1.31920695e+00] [ 6.19263506e+00] ..., [ 5.52571016e+04] [ 6.56975031e-01] [ 1.32541170e+01]] dh [[ -1.61431007e+01] [ 1.11354284e+01] [ -5.96233797e+00] ..., [ 5.50620898e+04] [ -1.52027397e+01] [ 2.52831030e+00]] dw [[ 3.35388470e+00] [ 1.31920695e+00] [ 6.19263506e+00] ..., [ 5.52571016e+04] [ 6.56975031e-01] [ 1.32541170e+01]] dh [[ -1.61431007e+01] [ 1.11354284e+01] [ -5.96233797e+00] ..., [ 5.50620898e+04] [ -1.52027397e+01] [ 2.52831030e+00]] dw [[ -7.99755469e+11] [ -1.12661050e+12] [ -4.95364628e+10] ..., [ -1.36656558e+15] [ 9.77799021e+11] [ 6.54851768e+11]] dh [[ 7.60489574e+11] [ 5.36720736e+11] [ -8.07536067e+10] ..., [ -1.39083013e+15] [ -9.40745359e+11] [ 1.08815345e+12]] dw [[ -7.99755469e+11] [ -1.12661050e+12] [ -4.95364628e+10] ..., [ -1.36656558e+15] [ 9.77799021e+11] [ 6.54851768e+11]] dh [[ 7.60489574e+11] [ 5.36720736e+11] [ -8.07536067e+10] ..., [ -1.39083013e+15] [ -9.40745359e+11] [ 1.08815345e+12]] dw [[ 1.79596433e+29] [ -6.17638350e+29] [ -1.23744272e+30] ..., [ inf] [ 1.12123926e+30] [ 4.85978886e+29]] dh [[ -9.36732242e+29] [ -5.96307005e+29] [ 7.51251514e+29] ..., [ inf] [ 1.17530431e+30] [ -9.80356859e+29]] /home/wu/faster_rcnn/py-faster-rcnn/tools/../lib/fast_rcnn/bbox_transform.py:56: RuntimeWarning: invalid value encountered in subtract pred_boxes[:, 0::4] = pred_ctr_x - 0.5 pred_w /home/wu/faster_rcnn/py-faster-rcnn/tools/../lib/fast_rcnn/bbox_transform.py:58: RuntimeWarning: invalid value encountered in subtract pred_boxes[:, 1::4] = pred_ctr_y - 0.5 pred_h /home/wu/faster_rcnn/py-faster-rcnn/tools/../lib/rpn/proposal_layer.py:175: RuntimeWarning: invalid value encountered in greater_equal keep = np.where((ws >= min_size) & (hs >= min_size))[0] dw [[ 1.79596433e+29] [ -6.17638350e+29] [ -1.23744272e+30] ..., [ inf] [ 1.12123926e+30] [ 4.85978886e+29]] dh [[ -9.36732242e+29] [ -5.96307005e+29] [ 7.51251514e+29] ..., [ inf] [ 1.17530431e+30] [ -9.80356859e+29]] dw [[ nan] [ nan] [ nan] ..., [ nan] [ nan] [ nan]] dh [[ nan] [ nan] [ nan] ..., [ nan] [ nan] [ nan]]

[meetps]

Check my comment here