`_skip_resize` ignored on detector inferenece

talcs commented 1 week ago

🐛 Describe the bug

Background

The FCOS object detector accepts **kwargs, of which is the _skip_resize flag, to be passed directly to GeneralizedRCNNTransform at FCOS init, here. If not specified, any input image should be resized according to the default parameters (min_size=800,max_size=1333).

Bug

However, in inference mode, even when passing _skip_resize=True when constructing FCOS, it is ignored by GeneralizedRCNNTransform, who will resize the image anyway. It can be clearly seen here.

The code in transform.py should be changed from this (only skipping resize at training):

if self.training:
    if self._skip_resize:
        return image, target
    size = self.torch_choice(self.min_size)
else:
    size = self.min_size[-1]
image, target = _resize_image_and_masks(image, size, self.max_size, target, self.fixed_size)

to this (skipping resize anyway):

if self._skip_resize:
    return image, target
if self.training:
    size = self.torch_choice(self.min_size)
else:
    size = self.min_size[-1]
image, target = _resize_image_and_masks(image, size, self.max_size, target, self.fixed_size)

Versions

Collecting environment information... PyTorch version: 1.12.1+cu116 Is debug build: False CUDA used to build PyTorch: 11.6 ROCM used to build PyTorch: N/A

OS: Ubuntu 22.04.3 LTS (x86_64) GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 Clang version: Could not collect CMake version: Could not collect Libc version: glibc-2.35

Python version: 3.10.12 (main, Sep 11 2024, 15:47:36) [GCC 11.4.0] (64-bit runtime) Python platform: Linux-5.15.0-107-generic-x86_64-with-glibc2.35 Is CUDA available: True CUDA runtime version: 11.5.119 CUDA_MODULE_LOADING set to: GPU models and configuration: Could not collect Nvidia driver version: Could not collect cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A Is XNNPACK available: True

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): 32 On-line CPU(s) list: 0-31 Vendor ID: GenuineIntel Model name: 13th Gen Intel(R) Core(TM) i9-13900K CPU family: 6 Model: 183 Thread(s) per core: 2 Core(s) per socket: 24 Socket(s): 1 Stepping: 1 CPU max MHz: 5800.0000 CPU min MHz: 800.0000 BogoMIPS: 5990.40 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 tsc_known_freq pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault ssbd ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid rdseed adx smap clflushopt clwb intel_pt sha_ni xsaveopt xsavec xgetbv1 xsaves split_lock_detect avx_vnni dtherm ida arat pln pts hwp hwp_notify hwp_act_window hwp_epp hwp_pkg_req umip pku ospke waitpkg gfni vaes vpclmulqdq tme rdpid movdiri movdir64b fsrm md_clear serialize pconfig arch_lbr flush_l1d arch_capabilities Virtualization: VT-x L1d cache: 896 KiB (24 instances) L1i cache: 1.3 MiB (24 instances) L2 cache: 32 MiB (12 instances) L3 cache: 36 MiB (1 instance) NUMA node(s): 1 NUMA node0 CPU(s): 0-31 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 Retbleed: Not affected Vulnerability Spec rstack overflow: Not affected 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 IBRS; IBPB conditional; RSB filling; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S Vulnerability Srbds: Not affected Vulnerability Tsx async abort: Not affected

Versions of relevant libraries: [pip3] numpy==1.26.1 [pip3] torch==1.12.1+cu116 [pip3] torchaudio==0.12.1+cu116 [pip3] torchvision==0.13.1+cu116 [conda] Could not collect

NicolasHug commented 2 days ago

Hi @talcs ,

Thanks for the report. It's a bit of an unconventional design, but that _skip_resize parameter is not meant to be used by users: it has a leading underscore which, for methods/attributes/functions means "private" in Python, and it's only available through kwargs, which makes it 'invisible' from the function's interface.

I do not remember the details here, but with all that, I suspect that parameter was never meant to be user-facing, and that's why it may only work in some specific cases, like when self.training is True.

May I ask why you would need this to be supported in inference mode as well?

talcs commented 2 days ago

Hey,

I have images that are much bigger than the detector's default and I don't want them downscaled, as all the small objects pretty much disappear.

Since FCOS is fully convolutional, it is completely scalable and adaptive to the input size.

On Wed, Nov 27, 2024, 17:47 Nicolas Hug @.***> wrote:

Hi @talcs https://github.com/talcs ,

Thanks for the report. It's a bit of an unconventional design, but that _skip_resize parameter is not meant to be used by users: it has a leading underscore which, for methods/attributes/functions means "private" in Python, and it's only available through kwargs, which makes it 'invisible' from the function's interface.

I do not remember the details here, but with all that, I suspect that parameter was never meant to be user-facing, and that's why it may only work in some specific cases, like when self.training is True.

May I ask why you would need this to be supported in inference mode as well?

— Reply to this email directly, view it on GitHub https://github.com/pytorch/vision/issues/8735#issuecomment-2504207616, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEIPDR6OVG4MC5LKPXF4QY32CXSPTAVCNFSM6AAAAABSD5VYGKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKMBUGIYDONRRGY . You are receiving this because you were mentioned.Message ID: @.***>

pytorch / vision