Update app.py
Browse files
app.py
CHANGED
@@ -64,39 +64,39 @@ use_fp16 = False
|
|
64 |
|
65 |
checkpoint_path = 'checkpoints/unival_s2_hs/checkpoint1.pt'
|
66 |
|
67 |
-
# Load ckpt & config for Image Captioning
|
68 |
-
caption_overrides={"eval_cider":False, "beam":5, "max_len_b":22, "no_repeat_ngram_size":3, "seed":7, "unnormalized": False,
|
69 |
-
|
70 |
-
|
71 |
-
caption_models, caption_cfg, caption_task = checkpoint_utils.load_model_ensemble_and_task(
|
72 |
-
|
73 |
-
|
74 |
-
)
|
75 |
-
|
76 |
-
# Load ckpt & config for Refcoco
|
77 |
-
refcoco_overrides = {"bpe_dir":"utils/BPE", "video_model_path": None}
|
78 |
-
|
79 |
-
refcoco_models, refcoco_cfg, refcoco_task = checkpoint_utils.load_model_ensemble_and_task(
|
80 |
-
|
81 |
-
|
82 |
-
)
|
83 |
-
refcoco_cfg.common.seed = 7
|
84 |
-
refcoco_cfg.generation.beam = 5
|
85 |
-
refcoco_cfg.generation.min_len = 4
|
86 |
-
refcoco_cfg.generation.max_len_a = 0
|
87 |
-
refcoco_cfg.generation.max_len_b = 4
|
88 |
-
refcoco_cfg.generation.no_repeat_ngram_size = 3
|
89 |
-
|
90 |
-
# Load pretrained ckpt & config for VQA
|
91 |
-
parser = options.get_generation_parser()
|
92 |
-
input_args = ["", "--task=vqa_gen", "--beam=100", "--unnormalized", f"--path={checkpoint_path}", "--bpe-dir=utils/BPE"]
|
93 |
-
args = options.parse_args_and_arch(parser, input_args)
|
94 |
-
vqa_cfg = convert_namespace_to_omegaconf(args)
|
95 |
-
vqa_task = tasks.setup_task(vqa_cfg.task)
|
96 |
-
vqa_models, vqa_cfg = checkpoint_utils.load_model_ensemble(
|
97 |
-
|
98 |
-
|
99 |
-
)
|
100 |
|
101 |
# Load pretrained ckpt & config for Generic Interface
|
102 |
parser = options.get_generation_parser()
|
@@ -110,17 +110,17 @@ general_models, general_cfg = checkpoint_utils.load_model_ensemble(
|
|
110 |
)
|
111 |
|
112 |
# move models to gpu
|
113 |
-
move2gpu(caption_models, caption_cfg)
|
114 |
-
move2gpu(refcoco_models, refcoco_cfg)
|
115 |
-
move2gpu(vqa_models, vqa_cfg)
|
116 |
move2gpu(general_models, general_cfg)
|
117 |
|
118 |
-
# Initialize generator
|
119 |
-
caption_generator = caption_task.build_generator(caption_models, caption_cfg.generation)
|
120 |
-
refcoco_generator = refcoco_task.build_generator(refcoco_models, refcoco_cfg.generation)
|
121 |
-
vqa_generator = vqa_task.build_generator(vqa_models, vqa_cfg.generation)
|
122 |
-
vqa_generator.zero_shot = True
|
123 |
-
vqa_generator.constraint_trie = None
|
124 |
general_generator = general_task.build_generator(general_models, general_cfg.generation)
|
125 |
|
126 |
# Construct image transforms
|
|
|
64 |
|
65 |
checkpoint_path = 'checkpoints/unival_s2_hs/checkpoint1.pt'
|
66 |
|
67 |
+
# # Load ckpt & config for Image Captioning
|
68 |
+
# caption_overrides={"eval_cider":False, "beam":5, "max_len_b":22, "no_repeat_ngram_size":3, "seed":7, "unnormalized": False,
|
69 |
+
# "bpe_dir":"utils/BPE", "video_model_path": None,}
|
70 |
+
|
71 |
+
# caption_models, caption_cfg, caption_task = checkpoint_utils.load_model_ensemble_and_task(
|
72 |
+
# utils.split_paths(checkpoint_path),
|
73 |
+
# arg_overrides=caption_overrides
|
74 |
+
# )
|
75 |
+
|
76 |
+
# # Load ckpt & config for Refcoco
|
77 |
+
# refcoco_overrides = {"bpe_dir":"utils/BPE", "video_model_path": None}
|
78 |
+
|
79 |
+
# refcoco_models, refcoco_cfg, refcoco_task = checkpoint_utils.load_model_ensemble_and_task(
|
80 |
+
# utils.split_paths(checkpoint_path),
|
81 |
+
# arg_overrides=refcoco_overrides
|
82 |
+
# )
|
83 |
+
# refcoco_cfg.common.seed = 7
|
84 |
+
# refcoco_cfg.generation.beam = 5
|
85 |
+
# refcoco_cfg.generation.min_len = 4
|
86 |
+
# refcoco_cfg.generation.max_len_a = 0
|
87 |
+
# refcoco_cfg.generation.max_len_b = 4
|
88 |
+
# refcoco_cfg.generation.no_repeat_ngram_size = 3
|
89 |
+
|
90 |
+
# # Load pretrained ckpt & config for VQA
|
91 |
+
# parser = options.get_generation_parser()
|
92 |
+
# input_args = ["", "--task=vqa_gen", "--beam=100", "--unnormalized", f"--path={checkpoint_path}", "--bpe-dir=utils/BPE"]
|
93 |
+
# args = options.parse_args_and_arch(parser, input_args)
|
94 |
+
# vqa_cfg = convert_namespace_to_omegaconf(args)
|
95 |
+
# vqa_task = tasks.setup_task(vqa_cfg.task)
|
96 |
+
# vqa_models, vqa_cfg = checkpoint_utils.load_model_ensemble(
|
97 |
+
# utils.split_paths(vqa_cfg.common_eval.path),
|
98 |
+
# task=vqa_task
|
99 |
+
# )
|
100 |
|
101 |
# Load pretrained ckpt & config for Generic Interface
|
102 |
parser = options.get_generation_parser()
|
|
|
110 |
)
|
111 |
|
112 |
# move models to gpu
|
113 |
+
# move2gpu(caption_models, caption_cfg)
|
114 |
+
# move2gpu(refcoco_models, refcoco_cfg)
|
115 |
+
# move2gpu(vqa_models, vqa_cfg)
|
116 |
move2gpu(general_models, general_cfg)
|
117 |
|
118 |
+
# # Initialize generator
|
119 |
+
# caption_generator = caption_task.build_generator(caption_models, caption_cfg.generation)
|
120 |
+
# refcoco_generator = refcoco_task.build_generator(refcoco_models, refcoco_cfg.generation)
|
121 |
+
# vqa_generator = vqa_task.build_generator(vqa_models, vqa_cfg.generation)
|
122 |
+
# vqa_generator.zero_shot = True
|
123 |
+
# vqa_generator.constraint_trie = None
|
124 |
general_generator = general_task.build_generator(general_models, general_cfg.generation)
|
125 |
|
126 |
# Construct image transforms
|