NA's for predictions using grayscale photo

[sitkensis22]

Describe the bug The predict function does not work with a black and white (i.e., grayscale) photo as expected. Instead of returning predictions, it returns NA's.

To Reproduce Code and steps to reproduce the behavior:

#remotes::install_github("maju116/platypus")

# set working directory
setwd(mypath)

# load packages
library(remotes)
library(platypus)
library(tidyverse)
library(abind)
library(reticulate)

# set conda environment
use_condaenv(conda_list()[4,2])

# setup yolo for grayscale photo
test_yolo_bw <- yolo3(
  net_h = 416, # Input image height. Must be divisible by 32
  net_w = 416, # Input image width. Must be divisible by 32
  grayscale = TRUE, # Should images be loaded as grayscale or RGB
  n_class = 80, # Number of object classes (80 for COCO dataset)
  anchors = coco_anchors # Anchor boxes
)
# inspect model
test_yolo_bw

# load darknet weights (note these must be downloaded)
test_yolo_bw %>% load_darknet_weights("yolov3.weights")

# store file path and name for photo
test_img_paths <- file.path(getwd(),"PDEP_2020-06-05_0303_0437.JPG") 

# load image and format array for analysis
test_imgs <- test_img_paths %>%
  map(~ {
    image_load(., target_size = c(416,416), color_mode = "grayscale") %>%
      image_to_array()  %>% `/`(255)}) %>%
  abind(along = 4)  %>%
  aperm(c(4, 1:3))

# get predictions of bounding boxes and class probabilities
test_preds <- test_yolo_bw %>% predict(test_imgs)

Expected behavior I expect the test_preds to be the predictions of bounding boxes and class probabilities: str(test_preds) List of 3 $ : num [1:4, 1:13, 1:13, 1:3, 1:85] 0.0249 0.0207 0.0028 0.0333 0.0526 ... $ : num [1:4, 1:26, 1:26, 1:3, 1:85] -0.03215 -0.03135 -0.0133 -0.05226 -0.00532 ... $ : num [1:4, 1:52, 1:52, 1:3, 1:85] 0.00578 0.00599 0.00452 0.01023 0.00315 ...

Screenshots If applicable, add screenshots to help explain your problem.

Session information (please complete the following information):

• OS: [e.g. iOS]:
• R version:
• Python version:
• TensorFlow (Python) version (tensorflow::tf_version()):
• R session information (sessionInfo()):

sessionInfo()
R version 4.2.2 (2022-10-31 ucrt)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 10 x64 (build 19045)

Matrix products: default

locale:
[1] LC_COLLATE=English_United States.utf8  LC_CTYPE=English_United States.utf8   
[3] LC_MONETARY=English_United States.utf8 LC_NUMERIC=C                          
[5] LC_TIME=English_United States.utf8    

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
 [1] reticulate_1.28   abind_1.4-5       lubridate_1.9.2   forcats_1.0.0    
 [5] stringr_1.5.0     dplyr_1.1.1       purrr_1.0.1       readr_2.1.4      
 [9] tidyr_1.3.0       tibble_3.2.1      ggplot2_3.4.2     tidyverse_2.0.0  
[13] platypus_0.1.1    keras_2.11.1      tensorflow_2.11.0

loaded via a namespace (and not attached):
 [1] progress_1.2.2     tidyselect_1.2.0   remotes_2.4.2      lattice_0.20-45   
 [5] colorspace_2.1-0   vctrs_0.6.1        generics_0.1.3     base64enc_0.1-3   
 [9] utf8_1.2.3         XML_3.99-0.14      rlang_1.1.0        pillar_1.9.0      
[13] glue_1.6.2         withr_2.5.0        rappdirs_0.3.3     RColorBrewer_1.1-3
[17] lifecycle_1.0.3    munsell_0.5.0      gtable_0.3.3       tzdb_0.3.0        
[21] tfruns_1.5.1       fansi_1.0.4        Rcpp_1.0.10        scales_1.2.1      
[25] jsonlite_1.8.4     gridExtra_2.3      hms_1.1.3          png_0.1-8         
[29] stringi_1.7.12     grid_4.2.2         rprojroot_2.0.3    here_1.0.1        
[33] cli_3.6.1          tools_4.2.2        magrittr_2.0.3     crayon_1.5.2      
[37] whisker_0.4.1      pkgconfig_2.0.3    zeallot_0.1.0      Matrix_1.5-1      
[41] prettyunits_1.1.1  timechange_0.2.0   rstudioapi_0.14    R6_2.5.1          
[45] compiler_4.2.2    

tensorflow::tf_version()
[1] ‘2.10’

python version is 3.7

Additional context I have no trouble running the example with a color photo, but with a grayscale photo I get this for the predictions: str(test_preds) List of 3 $ : num [1:6, 1:13, 1:13, 1:3, 1:85] NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN ... $ : num [1:6, 1:26, 1:26, 1:3, 1:85] NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN ... $ : num [1:6, 1:52, 1:52, 1:3, 1:85] NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN ...

Maybe the COCO training data did not include any grayscale photos?