In the baseline code type index is out of range

[ankana-001]

In the baseline code following code section

output_file = f'{OUT_DIR}/raw/file_map.dat'
output_info = joblib.load(output_file)

semantic_predictions = []
composition_predictions = []
for input_file, output_root in tqdm(output_info):
    img = cv2.imread(input_file)
    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    np_map = np.load(f'{output_root}.raw.0.npy')
    hv_map = np.load(f'{output_root}.raw.1.npy')
    tp_map = np.load(f'{output_root}.raw.2.npy')

    pred_map = process_segmentation(np_map, hv_map, tp_map)
    type_freqs = process_composition(pred_map)
    semantic_predictions.append(pred_map)
    composition_predictions.append(type_freqs)
semantic_predictions = np.array(semantic_predictions)
composition_predictions = np.array(composition_predictions)

I am getting this error

  0%|                                                  | 0/1018 [00:00<?, ?it/s]

---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
Cell In[27], line 14
     11 tp_map = np.load(f'{output_root}.raw.2.npy')
     13 pred_map = process_segmentation(np_map, hv_map, tp_map)
---> 14 type_freqs = process_composition(pred_map)
     15 semantic_predictions.append(pred_map)
     16 composition_predictions.append(type_freqs)

Cell In[25], line 21, in process_composition(pred_map)
     18 # ! not all types exist within the same spatial location
     19 # ! so we have to create a placeholder and put them there
     20 type_freqs = np.zeros(NUM_TYPES)
---> 21 type_freqs[type_freqs_[0]] = type_freqs_[1]
     22 return type_freqs

IndexError: index 250 is out of bounds for axis 0 with size 7

Because the type_map is giving out array numbers like these
([ 0, 250, 252, 253], dtype=int32) in place of 1-6.

[ankana-001]

any update on this issue ?

[mpf0909]

I have the same issue. Did you manage to find a fix?

[mpf0909]

Okay found a fix. Just copy the overlay_prediction_contours from the tiatoolbox v1.0.0 release...

def overlay_prediction_contours(canvas: np.ndarray, inst_dict: dict, draw_dot: bool = False, type_colours: dict = None, inst_colours: union[np.ndarray, Tuple[int]] = (255, 255, 0), line_thickness: int = 2):

overlay = np.copy((canvas))

if inst_colours is None:
    inst_colours = random_colors(len(inst_dict))
    inst_colours = np.array(inst_colours) * 255
    inst_colours = inst_colours.astype(np.uint8)
elif isinstance(inst_colours, tuple):
    inst_colours = np.array([inst_colours] * len(inst_dict))
elif not isinstance(inst_colours, np.ndarray):
    raise ValueError(
        f"`inst_colours` must be np.ndarray or tuple: {type(inst_colours)}"
    )
inst_colours = inst_colours.astype(np.uint8)

for idx, [_, inst_info] in enumerate(inst_dict.items()):
    inst_contour = inst_info["contour"]
    if "type" in inst_info and type_colours is not None:
        inst_colour = type_colours[inst_info["type"]][1]
    else:
        inst_colour = (inst_colours[idx]).tolist()
    cv2.drawContours(
        overlay, [np.array(inst_contour)], -1, inst_colour, line_thickness
    )

    if draw_dot:
        inst_centroid = inst_info["centroid"]
        inst_centroid = tuple([int(v) for v in inst_centroid])
        overlay = cv2.circle(overlay, inst_centroid, 3, (255, 0, 0), -1)
return overlay