comment

test.py

@@ -103,15 +103,18 @@ def test(data,
     jdict, stats, ap, ap_class, wandb_images = [], [], [], [], []
     for batch_i, (img, targets, paths, shapes) in enumerate(tqdm(dataloader, desc=s)):
         img = img.to(device, non_blocking=True)
-        img = img.half() if half else img.float()  # uint8 to fp16/32
-        img /= 255.0  # 0 - 255 to 0.0 - 1.0
+        if 1:
+            assert (img.shape[-1] == 640)
+        img = img.half() if half else img.float()
+        # uint8 to fp16/32
+        # img /= 255.0  # 0 - 255 to 0.0 - 1.0 c# already done inside dataloader
         targets = targets.to(device)
         nb, _, height, width = img.shape  # batch size, channels, height, width
 
         with torch.no_grad():
             # Run model
             t = time_synchronized()
-            out, train_out = model(img, augment=augment)  # inference and training outputs
+            out, train_out = model(img, augment=augment)  # inference(4 coordination, obj conf, cls conf ) and training outputs(un normalized coordination in yolo format and 3 scales diferent outputs) (2,2,80,80,7), (2,2,40,40,7)  : 640/8=40
             t0 += time_synchronized() - t
 
             # Compute loss
@@ -139,7 +142,7 @@ def test(data,
                 continue
 
             # Predictions
-            predn = pred.clone()
+            predn = pred.clone() # *xyxy, conf, cls in predn  [x y ,w ,h, conf, cls]
             scale_coords(img[si].shape[1:], predn[:, :4], shapes[si][0], shapes[si][1])  # native-space pred
 
             # Append to text file

train.py

@@ -393,7 +393,7 @@ def train(hyp, opt, device, tb_writer=None):
                 # Plot
                 if plots and ni < 10:
                     f = save_dir / f'train_batch{ni}.jpg'  # filename
-                    Thread(target=plot_images, args=(imgs, targets, paths, f), daemon=True).start()
+                    Thread(target=plot_images, args=(imgs, targets, paths, f, opt.input_channels), daemon=True).start()
                     # if tb_writer:
                     #     tb_writer.add_image(f, result, dataformats='HWC', global_step=epoch)
                     #     tb_writer.add_graph(torch.jit.trace(model, imgs, strict=False), [])  # add model graph
@@ -418,6 +418,7 @@ def train(hyp, opt, device, tb_writer=None):
                 results, maps, times = test.test(data_dict,
                                                  batch_size=batch_size * 2,
                                                  imgsz=imgsz_test,
+                                                 save_json=True,
                                                  model=ema.ema,
                                                  single_cls=opt.single_cls,
                                                  dataloader=testloader,
@@ -715,7 +716,15 @@ if __name__ == '__main__':
 
 
 """
+TODO
+Anchors,
+    hyp['anchor_t'] = 4 let the AR<=4 => TODO check if valid 
+    Ive reduced anchors to 2 per anchors: 2
+Sampler : torch_weighted : WeightedRandomSampler
+
+
 python train.py --workers 8 --device 'cpu' --batch-size 32 --data data/coco.yaml --img 640 640 --cfg cfg/training/yolov7.yaml --weights 'v7' --name yolov7 --hyp data/hyp.scratch.p5.yaml
 --workers 8 --device cpu --batch-size 32 --data data/tir_od.yaml --img 640 640 --cfg cfg/training/yolov7.yaml --weights 'v7' --name yolov7 --cache-images --hyp data/hyp.tir_od.tiny.yaml --adam --norm-type single_image_percentile_0_1
 --workers 8 --device cpu --batch-size 32 --data data/tir_od.yaml --img 640 640 --cfg cfg/training/yolov7-tiny.yaml --weights 'v7' --name yolov7 --cache-images --hyp data/hyp.tir_od.tiny.yaml --adam --norm-type single_image_percentile_0_1 --input-channels 1 --multi-scale
+--multi-scale training with resized image resolution not good for TIR
 """

train_aux.py

@@ -335,7 +335,7 @@ def train(hyp, opt, device, tb_writer=None):
         optimizer.zero_grad()
         for i, (imgs, targets, paths, _) in pbar:  # batch -------------------------------------------------------------
             ni = i + nb * epoch  # number integrated batches (since train start)
-            imgs = imgs.to(device, non_blocking=True).float() / 255.0  # uint8 to float32, 0-255 to 0.0-1.0
+            imgs = imgs.to(device, non_blocking=True).float() #/ 255.0  # uint8 to float32, 0-255 to 0.0-1.0
 
             # Warmup
             if ni <= nw:

utils/plots.py

@@ -111,7 +111,7 @@ def output_to_target(output):
     return np.array(targets)
 
 
-def plot_images(images, targets, paths=None, fname='images.jpg', names=None, max_size=640, max_subplots=16):
+def plot_images(images, targets, paths=None, fname='images.jpg', input_channels=3, names=None, max_size=640, max_subplots=16):
     # Plot image grid with labels
 
     if isinstance(images, torch.Tensor):
@@ -148,7 +148,14 @@ def plot_images(images, targets, paths=None, fname='images.jpg', names=None, max
         if scale_factor < 1:
             img = cv2.resize(img, (w, h))
 
-        mosaic[block_y:block_y + h, block_x:block_x + w, :] = img
+        if img.ndim > 2: # GL no permute
+            # Convert
+            mosaic[block_y:block_y + h, block_x:block_x + w, :] = img
+        else:
+            # img = img[np.newaxis,...] # unsqueeze
+            mosaic[block_y:block_y + h, block_x:block_x + w, :] = np.repeat(img[np.newaxis, :, :], 3, axis=0).transpose(1, 2, 0)
+
+
         if len(targets) > 0:
             image_targets = targets[targets[:, 0] == i]
             boxes = xywh2xyxy(image_targets[:, 2:6]).T