[BUG] BndBox flickers #257
Description
Describe the bug
When using cvcuda.bndbox/cvcuda.bndbox_into in Python multiple times, it flickers.
Example correct (expected) output:
Example flicker (notice how out of the 3 rectangles, only 2 are being properly drawn):
Example flicker (notice how out of the 3 rectangles, only 1 is being properly drawn):
2 more examples of flickering (notice how out of the 3 rectangles, only 1 is being properly drawn; 1 is missing completely & another only has "a few pixels going down"):
In my real project I've also encountered issues where 2 of the 4 sides of a rectangle would be gray instead of the specified color.
I've ran the following code on two machines.
Machine1 is using a NVIDIA GeForce RTX 3060 Laptop GPU bare-metal and on Machine2 I've tested both the Tesla T4 and the NVIDIA A2 inside Docker (a devcontainer).
I've tried using the _into variants of all the methods used in this example code, the behavior is the same. I've also used the default cuda stream (i.e. not passing stream to any of the cvcuda methods and not calling retain_primary_context or anything), the behavior is the same.
If I were to replace the multiple calls to cvcuda.bndbox with a single one, with multiple cvcuda.BndBoxI, the issue wouldn't reproduce, but in my real project, without major refactoring, I have to do it this way.
In my real project I'm using other operations as well (e.g. resize, composite, customcrop, copymakeborder, stack, convert) and they seem to worsen the issue somehow. If I also do inference using the tritonclient with CUDA shared memory (I copy the memory from a cvcuda tensor using In my real project I also encode the tensors using PyNvVideoCodec, but I don't think that's a problem, because if I dump the tensors to disk before encoding them (see tritonclient.utils.cuda_shared_memory.set_shared_memory_region_from_dlpack) I get even worse results, sometimes even black regions of the frames.tensor_to_png inside the code) they also look "corrupted".
EDIT: Ignore what I said about triton, it was an issue on my side.
Steps/Code to reproduce bug
from PIL import Image
import cupy as cp
import cv2
import cvcuda
import numpy as np
import pycuda.driver as cuda
import PyNvVideoCodec as nvc
counter = 0
def tensor_to_png(tensor) -> None:
global counter
gpu_mat = cv2.cuda.createGpuMatFromCudaMemory(
tensor.shape[1],
tensor.shape[2],
cv2.CV_8UC3,
tensor.cuda().__cuda_array_interface__["data"][0],
)
rgb_mat = gpu_mat.download()
bgr_mat = cv2.cvtColor(rgb_mat, cv2.COLOR_RGB2BGR)
cv2.imwrite(f"frames/{counter}.png", bgr_mat)
counter += 1
# Use this method if you don't have OpenCV built with CUDA support
def tensor_to_png_cupy(tensor) -> None:
global counter
cupy_tensor = cp.asarray(tensor.cuda())
if cupy_tensor.ndim == 4:
cupy_tensor = cp.squeeze(cupy_tensor, axis=0)
np_tensor = cp.asnumpy(cupy_tensor)
if np_tensor.shape[0] == 3:
np_tensor = np.transpose(np_tensor, (1, 2, 0))
img = Image.fromarray(np_tensor.astype(np.uint8), mode="RGB")
img.save(f"frames/{counter}.png")
counter += 1
def cvcuda_rectangle(
img,
bbox,
color,
thickness,
*,
stream=None,
) -> None:
fill_alpha = 0
if thickness == -1:
fill_alpha = 255
return cvcuda.bndbox(
img,
cvcuda.BndBoxesI(
[
[
cvcuda.BndBoxI(
bbox,
thickness,
# RGBA
(*color, 255),
(*color, fill_alpha),
)
]
]
),
stream=stream,
)
# On Machine2, I use a different GPU by setting CUDA_VISIBLE_DEVICES
device_id = 0
cuda_device = cuda.Device(device_id)
cuda_ctx = cuda_device.retain_primary_context()
cuda_ctx.push()
cuda_stream = cvcuda.Stream()
# I've also tried it without any of the above lines, the behavior is the same
# You can get the same exact video from https://file-examples.com/index.php/sample-video-files/sample-mp4-files/
demuxer = nvc.CreateDemuxer(filename="file_example_MP4_1920_18MG.mp4")
decoder = nvc.CreateDecoder(
codec=demuxer.GetNvCodecId(),
usedevicememory=True,
cudacontext=cuda_ctx.handle,
cudastream=cuda_stream.handle,
)
color_conversion = (
cvcuda.ColorConversion.YUV2RGB
if decoder.GetPixelFormat() == nvc.Pixel_Format.YUV444
else cvcuda.ColorConversion.YUV2RGB_NV12
)
for packet in demuxer:
for frame in decoder.Decode(packet):
tensor = cvcuda.as_tensor(cvcuda.as_image(frame.nvcv_image()))
nhwc_tensor = cvcuda.reformat(
tensor, cvcuda.TensorLayout.NHWC, stream=cuda_stream
)
rgb_tensor = cvcuda.cvtcolor(nhwc_tensor, color_conversion, stream=cuda_stream)
# Any coords should do
for rect in [
(436, 242, 289, 558),
(486, 292, 289, 558),
(616, 907, 423, 172),
]:
rgb_tensor = cvcuda_rectangle(
rgb_tensor, rect, (0, 255, 0), 3, stream=cuda_stream
)
# Or use tensor_to_png_cupy
tensor_to_png(rgb_tensor)Expected behavior
The user should be able to call BndBox multiple times on the same tensor, and all calls should result in tensors with complete rectangles.
Environment overview
- Environment location: Bare-metal on Machine1, Docker (devcontainer) on Machine2
- Method of cuDF install: not using cuDF; cvcuda was installed with pip
Environment details
Click here to see the environment details of Machine1
***OS Information***
DISTRIB_DESCRIPTION="Ubuntu 20.04.6 LTS"
Linux 5.15.0-139-generic #149~20.04.1-Ubuntu SMP Wed Apr 16 08:29:56 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux
***GPU Information***
Sun Jun 29 21:06:31 2025
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.183.01 Driver Version: 535.183.01 CUDA Version: 12.2 |
|-----------------------------------------+----------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+======================+======================|
| 0 NVIDIA GeForce RTX 3060 ... Off | 00000000:01:00.0 On | N/A |
| N/A 55C P8 16W / 60W | 83MiB / 6144MiB | 35% Default |
| | | N/A |
+-----------------------------------------+----------------------+----------------------+
+---------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=======================================================================================|
| 0 N/A N/A 1968 G /usr/lib/xorg/Xorg 24MiB |
| 0 N/A N/A 3720 G /usr/lib/xorg/Xorg 53MiB |
+---------------------------------------------------------------------------------------+
***CPU***
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
Address sizes: 48 bits physical, 48 bits virtual
CPU(s): 16
On-line CPU(s) list: 0-15
Thread(s) per core: 2
Core(s) per socket: 8
Socket(s): 1
NUMA node(s): 1
Vendor ID: AuthenticAMD
CPU family: 25
Model: 80
Model name: AMD Ryzen 9 5900HS with Radeon Graphics
Stepping: 0
Frequency boost: enabled
CPU MHz: 1300.000
CPU max MHz: 3300,0000
CPU min MHz: 1200,0000
BogoMIPS: 6588.09
Virtualization: AMD-V
L1d cache: 256 KiB
L1i cache: 256 KiB
L2 cache: 4 MiB
L3 cache: 16 MiB
NUMA node0 CPU(s): 0-15
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed: Not affected
Vulnerability Spec rstack overflow: Mitigation; safe RET, no microcode
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Retpolines; IBPB conditional; IBRS_FW; STIBP always-on; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr rdpru wbnoinvd cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip pku ospke vaes vpclmulqdq rdpid overflow_recov succor smca fsrm
***CMake***
/usr/bin/cmake
cmake version 3.31.2
***g++***
/usr/bin/g++
g++ (Ubuntu 13.1.0-8ubuntu1~20.04.2) 13.1.0
***nvcc***
/usr/local/cuda-12.9/bin/nvcc
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2025 NVIDIA Corporation
Built on Wed_Apr__9_19:24:57_PDT_2025
Cuda compilation tools, release 12.9, V12.9.41
Build cuda_12.9.r12.9/compiler.35813241_0
***Python***
Python 3.11.12
Click here to see the environment details of Machine2
DISTRIB_DESCRIPTION="Ubuntu 22.04.5 LTS"
Linux 5da4e91c4f64 5.4.0-216-generic #236-Ubuntu SMP Fri Apr 11 19:53:21 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux
***GPU Information***
Sun Jun 29 18:15:45 2025
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.230.02 Driver Version: 535.230.02 CUDA Version: 12.2 |
|-----------------------------------------+----------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+======================+======================|
| 0 Tesla T4 Off | 00000000:5E:00.0 Off | 0 |
| N/A 57C P0 30W / 70W | 624MiB / 15360MiB | 0% Default |
| | | N/A |
+-----------------------------------------+----------------------+----------------------+
| 1 NVIDIA A2 Off | 00000000:AF:00.0 Off | 0 |
| 0% 59C P0 27W / 60W | 6640MiB / 15356MiB | 0% Default |
| | | N/A |
+-----------------------------------------+----------------------+----------------------+
+---------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=======================================================================================|
+---------------------------------------------------------------------------------------+
***CPU***
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 46 bits physical, 48 bits virtual
Byte Order: Little Endian
CPU(s): 112
On-line CPU(s) list: 0-111
Vendor ID: GenuineIntel
Model name: Intel(R) Xeon(R) Gold 6238R CPU @ 2.20GHz
CPU family: 6
Model: 85
Thread(s) per core: 2
Core(s) per socket: 28
Socket(s): 2
Stepping: 7
BogoMIPS: 4400.00
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cdp_l3 invpcid_single intel_ppin ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm mpx rdt_a avx512f avx512dq rdseed adx smap clflushopt clwb intel_pt avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local dtherm ida arat pln pts pku ospke avx512_vnni md_clear flush_l1d arch_capabilities
Virtualization: VT-x
L1d cache: 1.8 MiB (56 instances)
L1i cache: 1.8 MiB (56 instances)
L2 cache: 56 MiB (56 instances)
L3 cache: 77 MiB (2 instances)
NUMA node(s): 2
NUMA node0 CPU(s): 0,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,32,34,36,38,40,42,44,46,48,50,52,54,56,58,60,62,64,66,68,70,72,74,76,78,80,82,84,86,88,90,92,94,96,98,100,102,104,106,108,110
NUMA node1 CPU(s): 1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39,41,43,45,47,49,51,53,55,57,59,61,63,65,67,69,71,73,75,77,79,81,83,85,87,89,91,93,95,97,99,101,103,105,107,109,111
Vulnerability Gather data sampling: Mitigation; Microcode
Vulnerability Itlb multihit: KVM: Mitigation: Split huge pages
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Mitigation; Clear CPU buffers; SMT vulnerable
Vulnerability Retbleed: Mitigation; Enhanced IBRS
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Enhanced / Automatic IBRS; IBPB conditional; RSB filling; PBRSB-eIBRS SW sequence; BHI SW loop, KVM SW loop
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Mitigation; TSX disabled
***CMake***
***g++***
/usr/bin/g++
g++ (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
***nvcc***
/usr/local/cuda/bin/nvcc
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Built on Tue_Aug_15_22:02:13_PDT_2023
Cuda compilation tools, release 12.2, V12.2.140
Build cuda_12.2.r12.2/compiler.33191640_0
***Python***
Python 3.11.13