# GroupNorm's small constant in the denominator __C.GROUP_NORM.EPSILON = 1e-5 # ---------------------------------------------------------------------------- # # MISC options # ---------------------------------------------------------------------------- # # Number of GPUs to use (applies to both training and testing) __C.NUM_GPUS = 1 # The mapping from image coordinates to feature map coordinates might cause # some boxes that are distinct in image space to become identical in feature # coordinates. If DEDUP_BOXES > 0, then DEDUP_BOXES is used as the scale factor # for identifying duplicate boxes. # 1/16 is correct for {Alex,Caffe}Net, VGG_CNN_M_1024, and VGG16 __C.DEDUP_BOXES = 1. / 16. # Clip bounding box transformation predictions to prevent np.exp from # overflowing # Heuristic choice based on that would scale a 16 pixel anchor up to 1000 pixels __C.BBOX_XFORM_CLIP = np.log(1000. / 16.) # Pixel mean values (BGR order) as a (1, 1, 3) array # We use the same pixel mean for all networks even though it's not exactly what # they were trained with # "Fun" fact: the history of where these values comes from is lost (From Detectron lol) __C.PIXEL_MEANS = np.array([[[102.9801, 115.9465, 122.7717]]]) # For reproducibility __C.RNG_SEED = 0 __C.ablation = 0
# Misc options # ---------------------------------------------------------------------------- # # Number of GPUs to use (applies to both training and testing) __C.NUM_GPUS = 1 # Use NCCL for all reduce, otherwise use muji # Warning: if set to True, you may experience deadlocks __C.USE_NCCL = False # The mapping from image coordinates to feature map coordinates might cause # some boxes that are distinct in image space to become identical in feature # coordinates. If DEDUP_BOXES > 0, then DEDUP_BOXES is used as the scale factor # for identifying duplicate boxes. # 1/16 is correct for {Alex,Caffe}Net, VGG_CNN_M_1024, and VGG16 __C.DEDUP_BOXES = 1 / 16. # Clip bounding box transformation predictions to prevent np.exp from # overflowing # Heuristic choice based on that would scale a 16 pixel anchor up to 1000 pixels __C.BBOX_XFORM_CLIP = np.log(1000. / 16.) # Pixel mean values (BGR order) as a (1, 1, 3) array # We use the same pixel mean for all networks even though it's not exactly what # they were trained with # "Fun" fact: the history of where these values comes from is lost __C.PIXEL_MEANS = np.array([[[102.9801, 115.9465, 122.7717]]]) # For reproducibility...but not really because modern fast GPU libraries use # non-deterministic op implementations __C.RNG_SEED = 3
# ---------------------------------------------------------------------------- # # MISC options # ---------------------------------------------------------------------------- # # Numer of refinement times __C.REFINE_TIMES = 3 # Number of GPUs to use (applies to both training and testing) __C.NUM_GPUS = 1 # The mapping from image coordinates to feature map coordinates might cause # some boxes that are distinct in image space to become identical in feature # coordinates. If DEDUP_BOXES > 0, then DEDUP_BOXES is used as the scale factor # for identifying duplicate boxes. # 1/16 is correct for {Alex,Caffe}Net, VGG_CNN_M_1024, and VGG16 __C.DEDUP_BOXES = 1. / 8. # Clip bounding box transformation predictions to prevent np.exp from # overflowing # Heuristic choice based on that would scale a 16 pixel anchor up to 1000 pixels __C.BBOX_XFORM_CLIP = np.log(1000. / 8.) # Pixel mean values (BGR order) as a (1, 1, 3) array # We use the same pixel mean for all networks even though it's not exactly what # they were trained with # "Fun" fact: the history of where these values comes from is lost (From Detectron lol) __C.PIXEL_MEANS = np.array([[[102.9801, 115.9465, 122.7717]]]) # For reproducibility __C.RNG_SEED = 3
# Misc options # ---------------------------------------------------------------------------- # # Number of GPUs to use (applies to both training and testing) __C.NUM_GPUS = 1 # Use NCCL for all reduce, otherwise use muji # Warning: if set to True, you may experience deadlocks __C.USE_NCCL = False # The mapping from image coordinates to feature map coordinates might cause # some boxes that are distinct in image space to become identical in feature # coordinates. If DEDUP_BOXES > 0, then DEDUP_BOXES is used as the scale factor # for identifying duplicate boxes. # 1/16 is correct for {Alex,Caffe}Net, VGG_CNN_M_1024, and VGG16 __C.DEDUP_BOXES = 1 / 16. # Clip bounding box transformation predictions to prevent np.exp from # overflowing # Heuristic choice based on that would scale a 16 pixel anchor up to 1000 pixels __C.BBOX_XFORM_CLIP = np.log(1000. / 16.) # Pixel mean values (BGR order) as a (1, 1, 3) array # We use the same pixel mean for all networks even though it's not exactly what # they were trained with # "Fun" fact: the history of where these values comes from is lost __C.PIXEL_MEANS = np.array([[[102.9801, 115.9465, 122.7717]]]) __C.GRAY_IMAGES = False __C.WHITEN_IMAGES = True __C.COLOR_NORMALIZE = False
__C.GROUP_NORM.EPSILON = 1e-5 # ---------------------------------------------------------------------------- # # MISC options # ---------------------------------------------------------------------------- # # Number of GPUs to use (applies to both training and testing) __C.NUM_GPUS = 1 # The mapping from image coordinates to feature map coordinates might cause # some boxes that are distinct in image space to become identical in feature # coordinates. If DEDUP_BOXES > 0, then DEDUP_BOXES is used as the scale factor # for identifying duplicate boxes. # 1/16 is correct for {Alex,Caffe}Net, VGG_CNN_M_1024, and VGG16 __C.DEDUP_BOXES = 1. / 16. # Clip bounding box transformation predictions to prevent np.exp from # overflowing # Heuristic choice based on that would scale a 16 pixel anchor up to 1000 pixels __C.BBOX_XFORM_CLIP = np.log(1000. / 16.) # Pixel mean values (BGR order) as a (1, 1, 3) array # We use the same pixel mean for all networks even though it's not exactly what # they were trained with # "Fun" fact: the history of where these values comes from is lost (From Detectron lol) __C.PIXEL_MEANS = np.array([[[102.9801, 115.9465, 122.7717]]]) # For reproducibility __C.RNG_SEED = 3